Search

The existence of glass defects seriously affects the quality of glass products. Traditional glass defect recognition methods based on convolutional neural networks (CNNs) suffer from a long training time and low recognition accuracy. By introducing convolutional auto-encoder (CAE) into a CNN, we propose an auto-encoding convolutional neural network, and pre-train the convolution kernel by CAE, thereby reducing the training time caused by randomly initializing the convolution kernel. At the same time, in order to cope with the over-fitting problem caused by a small sample of glass defect dataset, we employ a fuzzy support vector machine (FSVM) instead of the Softmax classifier to classify glass defects. Furthermore, we propose a multi-channel auto-encoding convolutional neural network model to deal with misidentification of inclusions and tumor type defects due to small differences in feature space. The model takes a defect image and its enhanced image as input, and averages the output of each channel as the final output, thus achieving accurate recognition of glass defects. Experimental results show that the glass defect recognition method based on auto-encoding convolutional neural network can greatly reduce the training time while maintaining the same classification accuracy rate. The glass defect recognition method based on multi-channel auto-encoding convolutional neural network achieves a recognition rate of 95% for both inclusion and tumor type defects. As a result, and the overall recognition rate is increased from 92.6% using a single channel to 97%. [ABSTRACT FROM AUTHOR]

In this present contribution, thermal conductivity ofethene (TCE) above the critical temperature has been studied. The present data cover the temperature range from 283.46 to 425.00 K and the pressure range from 0.1 to 100 MPa. In the present investigation, various network-based strategies, named as artificial neural network (ANN) optimized with two evolutionary algorithms, including genetic algorithm (GA) and differential evolution (DE), were developed for assessing tTCE in supercritical region. The most comprehensive source of data, including around 256 experimental points, was utilized for ANN modeling. Data index plot, scatter plot, relative deviation diagram and root mean square error (RMSE), and coefficient of determination (R2) as the statistical parameters were used in this examination to evaluate the comprehensiveness of the developed ANN model. Results indicate that the GA-ANN is more accurate than DE-ANN to predict TCE in supercritical region. Also, among optimization algorithms, GA has the largest ability for optimizing the ANN network modeling with the RMSE of 4.2966 and determination coefficient (R2) of 0.9640. [ABSTRACT FROM AUTHOR]

Online recognition of glass defects plays an important role in improving the quality of glass products. This paper presents a conjugate gradient neural network-based method, which combines phase map and the defective image for recognition of glass defects. The boundary coordinates of the connected defect region are calculated and used to extract the defect region in the defective image correspondingly. The piecewise linear gray-level transformation is designed to reduce the noise and to enhance the signal-to-noise ratio of the defective image. The second iteration segmentation based on gray range is applied to calculate the low and high thresholds, and the ternary-valued defective image is acquired. The seven features calculated by Hu invariant moment and four features extracted from the ternary-valued defective image are used as inputs of the conjugate gradient neural network to recognize the defect type. Experimental results show that the accuracy of the recognition reaches up to 93 %. Copyright © 2016 John Wiley & Sons, Ltd. [ABSTRACT FROM AUTHOR]