Segmented analysis of time-of-flight diffraction ultrasound for flaw detection in welded steel plates using extreme learning machines

Silva, Lucas C.; Filho, Eduardo F. Simas; Albuquerque, Maria C.S.; Silva, Ivan C.; Farias, Cláudia Teresa Teles

doi:10.1016/j.ultras.2019.106057

Cited by 30 publications

(4 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the literature exploring SVM models for classifying welding defects, wave parameters, and statistical features were extracted from both time-domain and frequency-domain signals to serve as input for the SVM models [22], [23], [24], [25], [26], [27]. This same methodology is applicable to shallow neural network models, which also use these extracted features as input [28], [29], [30], [31], [32]. In contrast, deep neural network (DNN) models can directly use either time-domain signals or frequency spectra as model input [33], [34], [35], [36].…”

Section: Literature Reviewmentioning

confidence: 99%

Assessment of Machine Learning for Ultrasonic Nondestructive Evaluation of Alkali–Silica Reaction in Concrete

Sun,

Sabatino

2024

View full text Add to dashboard Cite

Alkali-silica reaction (ASR) is a type of material degradation in concrete structures that leads to concrete cracking and rebar corrosion, thereby reducing the material's structural integrity and the overall structure's lifetime and raising safety concerns. Ultrasonic nondestructive evaluation (NDE) has been proven to be a valuable technique for assessing concrete properties and monitoring ASR progression in concrete. However, the deployment and analysis of ultrasonic NDE and its data requires specialized expertise, often relying on the engineer's subjective interpretation. With the surge in computational power, artificial intelligence (AI) and machine learning (ML) algorithms have become popular in automating NDE data analysis. Various industrial sectors are increasingly adopting ML algorithms for NDE data analysis with a growing emphasis on AI-assisted automation. Regulatory agencies are also preparing for this technological shift, anticipating corresponding revisions in standards. Thus, there is an urgent need to identify the capabilities and limitations of current ML technologies for the evaluation of concrete material properties and damage status. Furthermore, the effects of various factors on ML model performance must be thoroughly investigated.The study summarized herein evaluated the effectiveness of two ML models (i.e., support vector regression (SVR) and deep neural network (DNN)) in predicting concrete material damage induced by ASR based on the long-term ultrasonic monitoring data. Four distinct concrete specimens were cast with artificially induced ASR, and over a period exceeding 500 days, ultrasonic signals and expansion data were continuously collected. For the SVR model, wave velocity and 12 other wave features were extracted from the ultrasonic signals, with 6 out of 13 features selected as input for the model. Different combinations of training and testing datasets were designed to explore factors influencing prediction performance, including the range of data within training and testing sets, in addition to various signal preprocessing methodologies. These findings suggest the importance of using a training dataset with a broader data range compared with testing datasets for improved model performance alongside consistent signal preprocessing across datasets.Additionally, this work studied the effect of temperature on model performance, revealing significant prediction errors when the temperature of testing data differed from that of training data. Furthermore, this study examined training and testing datasets from two distinct specimen batches, revealing the SVR model's superior generalization ability compared with DNN when considering data from different batches. Lastly, DNN models were trained and tested using both time-domain signals and frequency spectra. The results showed potential for achieving high regression performance (e.g., the ML model was able to predict ASR expansion with relatively high accuracy), although further efforts in model tuning and data preprocessing are still requir...

show abstract

Section: Literature Reviewmentioning

confidence: 99%

Assessment of Machine Learning for Ultrasonic Nondestructive Evaluation of Alkali–Silica Reaction in Concrete

Sun,

Sabatino

2024

View full text Add to dashboard Cite

show abstract

“…Optical sensors are most commonly used for process optimization; photodiodes and cameras observe vapor plumes at UV-visible frequencies, while thermal radiation and its distribution are observed in the IR-range by spectrometers and pyrometers [3][4][5][6][7]. Other automated techniques include acoustic emission [8,9], ultrasound [10,11], and X-ray radiography [12]. Because these individual technologies have specific competencies and limitations in identifying particular types of defect, state-of-theart defect-detection utilizes multiple-sensor fusion in order to complement the capabilities of specific sensors [13,14].…”

Section: Current Trends In Industrial Joiningmentioning

confidence: 99%

Laser Doppler Vibrometry for Evaluating the Quality of Welds in Lithium-Ion Supercells

Ratner

Wood

Chowanietz³

et al. 2022

Energies

View full text Add to dashboard Cite

The inspection of the quality of welds in battery packs plays an important role in ensuring safety during the manufacturing and operation of energy-storage devices in automotive vehicles during service. This research investigated the novel application of laser Doppler vibrometry, a widely used non-destructive optical technique for modal analysis, to the post-weld evaluation of micro-TIG-welded interconnections in lithium-ion supercells. The experimental modal analysis showed features in the modal models of the supercells that were unique to their welding conditions. The comparisons between the supercells showed an absence of linear correlations between the modal parameters and the welding current, as well as differences in the welding parameters obtained from the negative and positive terminals of the cylindrical cells. These findings suggested that the modal parameters of the supercells were more strongly influenced by the rigidity of the structural materials than by the localized compliance of the welded interconnections. While this investigation demonstrated a method for using laser Doppler vibrometry to distinguish between different welding conditions in lithium-ion supercells at a structural level, further development is needed to identify the weld quality of individual interconnections.

show abstract

“…Huang et al [15] designed an eddy current orthogonal axial probe for eddy current detection of carbon steel plate welds, and the experiment proved that the probe can effectively detect the effect of unevenness of the weld surface on the lifting-off effect. Silva et al proposed a Segmented analysis of timeof-flight diffraction ultrasound for flaw model, which is able to identify the type of defects in the welding process by analyzing ultrasonic signal fragments [23]. Du et al [5] proposed a fusion of background subtraction and grayscale wave analysis for defect classification by analyzing weld x-ray.…”

Section: Introductionmentioning

confidence: 99%

WeldNet: a lightweight deep learning model for welding defect recognition

Wang,

et al. 2024

Weld World

View full text Add to dashboard Cite

Weld defect detection is an important task in the welding process. Although there are many excellent weld defect detection models, there is still much room for improvement in stability and accuracy. In this study, a lightweight deep learning model called Weld-Net is proposed to improve the existing weld defect recognition network for its poor generalization performance, overfitting and large memory occupation, using a design with a small number of parameters but with better performance. We also proposed ensembledistillation strategy in the training process, which effectively improved the accuracy rate and proposed an improved model ensemble scheme. The experimental results show that the final designed WeldNet model performs well in detecting weld defects and achieves state-of-the-art performance. Its number of parameters is only 26.8% of that of ResNet18, but the accuracy is 8.9% higher, while achieving 41 FPS on cpu to meet the demand of real-time operation. The study is of guiding significance for solving practical problems in weld defect detection, and provides new ideas for the application of deep learning in industry. The code used in this article is available at: https://github.com/Wanglaoban3/WeldNet.git

show abstract

Segmented analysis of time-of-flight diffraction ultrasound for flaw detection in welded steel plates using extreme learning machines

Cited by 30 publications

References 21 publications

Assessment of Machine Learning for Ultrasonic Nondestructive Evaluation of Alkali–Silica Reaction in Concrete

Assessment of Machine Learning for Ultrasonic Nondestructive Evaluation of Alkali–Silica Reaction in Concrete

Laser Doppler Vibrometry for Evaluating the Quality of Welds in Lithium-Ion Supercells

WeldNet: a lightweight deep learning model for welding defect recognition

Contact Info

Product

Resources

About