In this study, neat polyacrylonitrile (PAN) and fumed silica (FS)-doped PAN membranes (0.1, 0.5 and 1 wt% doped PAN/FS) are prepared using the phase inversion method and are characterised extensively. According to the Fourier Transform Infrared (FTIR) spectroscopy analysis, the addition of FS to the neat PAN membrane and the added amount changed the stresses in the membrane structure. The Scanning Electron Microscope (SEM) results show that the addition of FS increased the porosity of the membrane. The water content of all fabricated membranes varied between 50% and 88.8%, their porosity ranged between 62.1% and 90%, and the average pore size ranged between 20.1 and 21.8 nm. While the neat PAN membrane’s pure water flux is 299.8 L/m2 h, it increased by 26% with the addition of 0.5 wt% FS. Furthermore, thermal gravimetric analysis (TGA) and differential thermal analysis (DTA) techniques are used to investigate the membranes’ thermal properties. Finally, the mechanical characterisation of manufactured membranes is performed experimentally with tensile testing under dry and wet conditions. To be able to provide further explanation to the explored mechanics of the membranes, numerical methods, namely the finite element method and Mori–Tanaka mean-field homogenisation are performed. The mechanical characterisation results show that FS reinforcement increases the membrane rigidity and wet membranes exhibit more compliant behaviour compared to dry membranes.

No abstract

Achieving satisfactory results with Convolutional Neural Networks (CNNs) depends on how effectively the filters are trained. Conventionally, an appropriate number of filters is carefully selected, the filters are initialized with a proper initialization method and trained with backpropagation over several epochs. This training scheme requires a large labeled dataset, which is costly and time-consuming to obtain. In this study, we propose an unsupervised approach that extracts convolutional filters from a given dataset in a self-organized manner by processing the training set only once without using backpropagation training. The proposed method allows for the extraction of filters from a given dataset in the absence of labels. In contrast to previous studies, we no longer need to select the best number of filters and a suitable filter weight initialization scheme. Applying this method to the MNIST, EMNIST-Digits, Kuzushiji-MNIST, and Fashion-MNIST datasets yields high test performances of 99.19%, 99.39%, 95.03%, and 90.11%, respectively, without applying backpropagation training or using any preprocessed and augmented data.INDEX TERMS Convolutional neural networks, feature extraction, unsupervised learning.

No abstract

No abstract