No abstract

Weld defect classification in radiographic images using unified deep neural network with multi-level features Cite this article as: Lu Yang and Hongquan Jiang, Weld defect classification in radiographic images using unified deep neural network with multi-level features, Journal of Intelligent Manufacturing

No abstract

To improve the accuracy of automatic defect classification, a novel algorithm has been developed based on the principal component analysis (PCA) and support vector machine (SVM) methods. The original defect data are transformed to principal component space by the PCA algorithm and then the optimal dataset is selected. Then, the SVM is used for defect classification. For estimating the actual classification accuracy of the proposed method in a concrete system, the bootstrap method is introduced. The experimental result demonstrates that the accuracy of the new method is 90.75%, which promotes the evaluation accuracy by 3.24% and 4.93% compared with the SVM and MLP-ANN, respectively. Furthermore, the new method takes less computing time than the MLP-ANN method.

Aiming to solve the problems of low accuracy of multi-step prediction and difficulty in determining the maximum number of prediction steps of chaotic time series, a multi-step time series prediction model based on the dilated convolution network and long short-term memory (LSTM), named the dilated convolution-long short-term memory (DC-LSTM), is proposed. The dilated convolution operation is used to extract the correlation between the predicted variable and correlational variables. The features extracted by dilated convolution operation and historical data of predicted variable are input into LSTM to obtain the desired multi-step prediction result. Furthermore, cross-correlation analyses (CCA) are applied to calculate the reasonable maximum prediction steps of chaotic time series. Actual applications of multi-step prediction were studied to demonstrate the effectiveness of the proposed model which has superiorities in RMSE, MAE and prediction accuracy because of the extraction of correlation between the predicted variable and correlational variables. Moreover, the proposed DC-LSTM model provides a new method for prediction of chaotic time series and lays a foundation for scientific data analysis of chaotic time series monitoring systems.

Analysis of the multivariable coupling relationship, detection of the features of the coupling, and quantification of different degrees of each variable for the coupling are important foundations for information modeling, key point identification, and fault tracing of complex electromechanical systems. It is significant to understand the reasons associated with system conditions exchange, and improve the abilities of accident prevention and safety control of the system. In order to study the multifractal properties of the multivariable coupling relationship of the production system in the process industry, coupling detrended fluctuation analysis (CDFA) was applied. The strength and sources of the multifractality were estimated by shuffling and phase randomization with confidence bands. Different degrees of each variable were qualitatively and quantitatively quantified by the Chi square test. Empirical results showed that the CDFA was suitable for analyzing the multivariable coupling relationship of complex electromechanical system, and the coupling exhibited an obvious multifractal feature. Long-range correlation and fat-tailed probability density function of variables are sources for the multifractality of the multivariable coupling relationship of production system in process industry. Some invariant values, such as the multifractality power and ranks of different degree of single variable for the coupling, were detected for a specific operation condition, which were meaningful for conditions identification and information quality control of complex electromechanical system in process industry.

Weld defect recognition plays an important role in the manufacturing process of large-scale equipment. Traditional methods generally include several serial steps, such as image preprocessing, region segmentation, feature extraction, and type recognition. The results of each step have significant impact on the accuracy of the final defect identification. The convolutional neural network (CNN) has strong pattern recognition ability, which can overcome the above problem. However, there are two problems: one is that the pooling strategy has poor dynamic adaptability, and the other is the insufficient feature selection ability. To overcome these problems, we propose a CNN-based weld-defect recognition method, which includes an improved pooling strategy and an enhanced feature selection method.According to the characteristics of the weld-defect image, an improved pooling strategy that considers the distribution of the pooling region and feature map is introduced. Additionally, in order to enhance the feature selection ability of the CNN, an enhanced feature selection method integrating the ReliefF algorithm with the CNN is proposed. A case study is presented for demonstrating the proposed techniques.The results show that the proposed method has higher accuracy than the traditional CNN method, and establish that the proposed CNN-based method is successfully applied for weld-defect recognition.

No abstract