On-line Welding Quality Monitoring Research Articles

EfficientNet-ECA: A lightweight network based on efficient channel attention for class-imbalanced welding defects classification

Welding defects recognition is crucial for ensuring weldment quality in robot arc welding. Nevertheless, due to unbalanced data distributions, limited computing resources in factory, as well as intraclass variability and interclass similarity among different welding defects, it is difficult to extract the most discriminative defect features from welding molten pool images on site, resulting in weak class-imbalanced defect recognition performance. To address the above issue, a novel lightweight network named EfficientNet-ECA is proposed for recognizing and classifying welding defects according to molten pool images of robot arc welding. Firstly, the EfficientNet-ECA network based on Efficient Channel Attention (ECA) is designed to extract the most discriminative features of different defects from molten pool images, where ECA is employed to enhance cross-channel information interactions. Secondly, a dynamic equalized focal loss function is proposed to both rebalance the loss contribution of class-imbalanced data of different defects and improve defects recognition accuracy. Subsequently, a class-imbalanced dataset containing eight typical welding defects is constructed to evaluate the EfficientNet-ECA model. Finally, the proposed model is comprehensively analyzed through ablation studies and compared with the existing state-of-the-art lightweight models, and results show that the proposed method exhibits better effectiveness and generalizability in classifying class-imbalanced defects across different welding scenarios, achieving the highest accuracy of 95.84% on a self-constructed AL5083 dataset and 96.50% on a publicly available SS304 dataset. Moreover, with fewer parameters and less complexity, the proposed method is more suitable for online welding quality monitoring in factory.

AF-FTTSnet: An end-to-end two-stream convolutional neural network for online quality monitoring of robotic welding

Online welding quality monitoring (WQM) is crucial for intelligent welding, and deep learning approaches considering spatiotemporal features for WQM tasks show great potential. However, one of the important challenges for existing approaches is to balance the spatiotemporal representation learning capability and computational efficiency, which makes it challenging to adapt welding processes with complex and drastic molten pool dynamic behavior. This paper proposes a novel approach for WQM using molten pool visual sensing and deep learning considering spatiotemporal features, the proposed deep learning network called attention fusion based frame-temporality two-stream network (AF-FTTSnet). Firstly, a passive vision sensor is used to acquire continuous dynamic molten pool images. Meanwhile, temporal difference images are computed to provide novel features and temporal representations. Then, a two-stream feature extraction module is designed to concurrently extract rich spatiotemporal features from molten pool images and temporal difference images. Finally, an attention fusion module with the ability to automatically identify and weight the most relevant features is designed to achieve optimal fusion of the two-stream features. The shop welding experimental results indicate that the proposed AF-FTTSnet model can effectively and robustly recognize five typical welding states during helium arc welding, with an accuracy of 99.26%. This model has been demonstrated to exhibit significant performance improvements compared to mainstream temporal sequence models. Available: https://github.com/Just199806/TSCNN/tree/master.

Online evaluation of resistance spot welding quality and defect classification

Resistance spot welding (RSW) is widely used in industrial product manufacturing, and weld quality is critical to ensure the safety and reliability of mechanical structures. With product quality improvement, the online monitoring of weld quality has become an urgent issue. Therefore, a method for online evaluation of RSW quality and automatic classification of welding defects based on mild steel is proposed. By acquiring the welding current and electrode voltage signals of the welding process in real time, a finite impulse response low-pass filter based on the Blackman window function is designed to perform zero-phase filtering of the dynamic resistance (DR). The influence of three welding parameters on the nugget diameter and shear strength of spot welding of DC01 mild steel was studied through the control variable method to determine the optimum welding parameters. Since the main defects of the RSW process are classified as expulsion defects and incomplete fusion defects, minor adjustments were made to the welding parameters to produce expulsion and incomplete fusion defects. Features in the time domain are extracted from the DR curves signals of normal, expulsion, and incomplete fusion welding spots, and the performance of the extracted features in the classifier is analyzed. Improvements are made to classify normal and expulsion spots with low accuracy, a new method is proposed that can accurately classify the quality of the welding spots. Through online evaluation tests of quality on 200 welding spots, the online evaluation test showed results that closely matched the validation test results. The results of the classification performance tests show that this study can be used for online evaluation of RSW quality and automatic classification of the welding spot defects of the DC01.

Prediction of weld back width based on top vision sensing during laser-MIG hybrid welding

Online welding quality monitoring has gained increasing attention in modern automatic production. This work proposed a prediction method of weld back width based on top vision during laser-MIG (Metal Inert-Gas) hybrid welding. A high-speed photography system was used to monitor the laser-MIG hybrid welding, which could observe the visual information from top of the weldment. Visual information including the features of keyhole, arc and molten pool, were extracted as model inputs by image processing and signal processing, and the weld back width was extracted as model output by 3D (Three-Dimension) point cloud processing. An ATT-LSTM (Attention - Long Short Term Memory) prediction model was constructed to predict the weld back width. Experimental results show that ATT-LSTM can not only improve the accuracy and generalization ability of weld back width prediction but also deliver interpretable analysis, compared with other methods such as RNN (Recurrent Neural Network), LSTM, GRU (Gated Recurrent Unit), LSTM-ATT, which MSE (mean square error) was 0.0848 mm2, R2 (r-squared) was 0.5900, CORR (correlation coefficient) was 0.7694 and the comprehensive prediction error (2σ) was 0.5853 mm. The most attention was focused on the keyhole feature, and the feature information reflecting droplet transfer and laser-arc coupling stability was also discussed.

A study on welding quality for the automatic vertical-position welding process based on Mahalanobis Distance method

Purpose: The welding quality and reducing production cost could be achieved by developing the automatic on-line welding quality monitoring system. However, investigation of welding fault to quantify the welding quality on the horizontal-position welding has been concentrated. Therefore, MD (Mahalanobis Distance) method on the vertical-position welding process by analysing the transform arc voltage and welding current gained from the on-line monitoring system has been applied. Design/methodology/approach: The transformed welding current and arc voltage data were taken from the experiment whereby the data number was 2500 data/s. The prediction of Contact Tip to Work Distance (CTWD) to gain best welding quality using the waveform variations were then taken from the experimental results. MD was employed to quantify the welding quality by analysing the transformed arc voltage and welding current. Finally, the optimal CTWD setting has verified the developed algorithms through additional experiments. Two kinds of experiments has been carried out by changing welding parameters artificially to verify the sensitivity and feasibility of WQ (Welding Quality) based on the concepts of MD and normal distribution. Findings: The results represented that WQ was fully capable of quantifying and qualifying the welding faults for automatic vertical-position welding process. Research limitations/implications: The arc welding process on the vertical-position compared to a horizontal-position welding is much more difficult because the metal transfer is influenced by the gravity force. To solve the problem, a new algorithm to monitor and control the welding fault during the arc welding process has been developed. Furthermore, optimization of welding parameters for the vertical-position welding process was really difficult to use the developed algorithms because they are only useful in selecting stored data and not for evaluating the effect of the variation of welding parameters on the weld ability. Practical implications: The developed algorithm could be achieved the highest welding quality at 15mm CTWD setting which the welding quality is 99.50% for the start section and 99.68% at the middle section. Originality/value: This paper proposed a new algorithm which employed the concepts of MD (Mahalanobis Distance) and normal distribution to describe a good quality welding.

Online welding quality monitoring for large-size electrical contact high frequency induction brazing

In the low voltage apparatus manufacturing industry, the welding quality of electrical contact directly determines the switching capacity, lifespan and reliability of electrical equipment. Currently, only a little research on online monitoring of large-size electrical contact high frequency induction brazing has been carried out. In this research, CJ400 electrical contact was used as samples to be welded in the way of pulse induction brazing. And the electrical parameters during the whole welding process were collected and integrated with the dual 32-bit ARM (Advanced RISC Machines) controller. The relationship between welding quality and integral value of electrical parameter was analyzed. The results showed that the deviation between the ideal integral value of electrical parameters and the actual integral value of electrical parameters could be identified an indication for welding quality.

Multisensor-based real-time quality monitoring by means of feature extraction, selection and modeling for Al alloy in arc welding

Multisensory data fusion-based online welding quality monitoring has gained increasing attention in intelligent welding process. This paper mainly focuses on the automatic detection of typical welding defect for Al alloy in gas tungsten arc welding (GTAW) by means of analzing arc spectrum, sound and voltage signal. Based on the developed algorithms in time and frequency domain, 41 feature parameters were successively extracted from these signals to characterize the welding process and seam quality. Then, the proposed feature selection approach, i.e., hybrid fisher-based filter and wrapper was successfully utilized to evaluate the sensitivity of each feature and reduce the feature dimensions. Finally, the optimal feature subset with 19 features was selected to obtain the highest accuracy, i.e., 94.72% using established classification model. This study provides a guideline for feature extraction, selection and dynamic modeling based on heterogeneous multisensory data to achieve a reliable online defect detection system in arc welding.

Analysis of an Air Borne Acoustic Signatures from Welding Process Using Empirical Mode Decomposition

This work present the application of Empirical Mode Decomposition (EMD) to analyze the air borne acoustic signatures commenced during welding process. In order to achieve goals, bead-on plate welding was done onto the carbon steel specimen using Metal Inert Gas Welding. At the same moment, the microphone with operating frequency of 31.5 Hz to 8 kHz used to collect air borne acoustic signatures. In analysis part, Empirical Mode Decomposition has been applied to the acoustic signals and the selected Intrinsic Mode Function (IMF) was presented in frequency-distance plot using spectrogram. Based on the analysis results, there were 3 significant IMF has been found. Those were IMF mode 3, mode 5 and mode 8 which lie within the frequency of 1500 Hz to 4500 Hz, 200 Hz to 800 Hz, and 40 Hz to 80 Hz respectively. The frequency-distance plot from spectrogram of IMF mode 3 showed a significant pattern which can be related with the discontinuity of welding. The discontinuity appears wherever low amplitude power detected in the frequency-distance plot of IMF mode 3. Moreover, the frequency-distance plot of IMF mode 5 and mode 8 can be significantly related with the spatter and weld pool oscillation condition. High power amplitude in frequency-distance plot of IMF mode 8 can indicates the offset of weld pool oscillation frequency and cause the existence of higher amount of spatter which resulting the high power amplitude in frequency-distance plot of IMF mode 5. In summary, it can be conclude that the application of EMD in the analysis of air borne acoustic signatures allow the detection of several phenomena in welding process which might lead to defect once in a time. This was found to be significant in the process of developing the online welding quality monitoring.

Online welding quality monitoring based on feature extraction of arc voltage signal

Arc sensing plays a significant role in the control and monitoring of welding quality for aluminum alloy pulsed gas touch argon welding (GTAW). A method for online quality monitoring was proposed based on the analysis of acquired arc voltage signal, through which two algorithms of feature extraction were developed in time and frequency domain, respectively. In time domain,the wavelet packet transform was carried out to eliminate the pulse interference of the feature parameter curve. In frequency domain, the other new algorithm was proposed based on the voltage power spectrum density (PSD) which was calculated by using the improved Welch algorithm and divided into five frequency bands before the statistic parameters were extracted. The correlation between the feature parameters in different frequency bands and welding defects were carefully analyzed to select a more sensitive one as the monitoring parameters. The proposed algorithms on this paper were verified to be capable of detecting lack of penetration, burn through, and the defect caused by lack of gas.

Feasibility study of Hierarchical Temporal Memories applied to welding diagnostics

Defect classification in on-line welding quality monitoring systems is an active area of research with a significant relevance to several industrial sectors where welding processes are extensively employed. Approaches based on some artificial intelligence implementations, like Artificial Neural Networks or Fuzzy Logic have been attempted, but their impact in real industrial scenarios is nowadays rather modest. In this paper a new approach based on Hierarchical Temporal Memories and the acquired plasma spectra is explored and analyzed by means of several arc-welding experimental tests. Results show the ability of the proposed solution to perform a suitable classification among several weld perturbations. The search for an optimal configuration of the algorithm and the usefulness of both spatial (spectral) and temporal identification of patterns will be also discussed, and the results will be compared with those provided by a solution based on feature selection and neural networks, exhibiting the better performance of the HTM model in terms of performance and handling of the input data.

On use of weld zone temperatures for online monitoring of weld quality in friction stir welding of naturally aged aluminium alloys

6063-T4 aluminium alloy sheets of 3 and 6mm thicknesses were friction stir butt welded using a square tool pin at a wide range of tool rotational speeds. Properties of obtained welds were characterised using tensile tests, optical micrographs, X-ray diffraction, and transmission electron microscopy. Shape, size, and distribution of precipitates in weld zones, and strength and ductility of welds were seen to directly correlate with peak temperatures in weld nugget and heat affected zones, independent of sheet thickness. In addition, fluctuations in measured temperature profiles, for 3mm sheets, were seen to correlate with an increase in scatter of weld nugget zone properties for 3mm sheets. Optimal weld strength and ductility were obtained for peak weld nugget zone temperatures of around 450°C and corresponding peak heat affected zone temperatures of around 360–380°C. Results obtained suggest that, at least for naturally aged aluminium alloys, nature of temperature evolution and magnitudes of peak temperatures in weld nugget and heat affected zones provide information on uniformity of properties in weld zones, overaging of heat affected zones, and formation of tunnel defects from improper material mixing at low weld zone temperatures.

Spectroscopic analysis of the plasma continuum radiation for on-line arc-welding defect detection

When plasma optical spectroscopy is applied to on-line welding quality monitoring, the plasma electronic temperature is commonly selected as the spectroscopic parameter to determine. However, several processing stages have to be considered in this case, including plasma emission line identification, which is significantly costly in terms of computational performance. In this paper, the wavelength associated with the maximum intensity of the plasma background emission is proposed as the new monitoring signal, as it will be demonstrated that there is a clear correlation between this parameter and the welding quality. The resulting processing scheme is clearly simpler, and experimental and field tests will prove the feasibility of the proposed technique.

On line monitoring of burning through for short circuit CO2 arc welding based on the self-organize feature map neural networks

A method for automatic detection of burning through of short circuit CO2 arc welding is presented. It is based on the extraction of arc signal features as well as classification of the obtained features using self-organize feature map(SOM) neural networks in order to get the weld quality information, for example, to determine if there is defect in the product. This is important for the on-line monitoring of weld quality especially in robotic welding and lay the foundation for the further real-time control of weld quality.