Optimization and prediction of ultra-fine glass fiber felt process parameters based on artificial neural network

Abstract

Ultra-fine glass fiber felt (fiber diameter ⩽3 μm) is prepared by the flame blowing process with superior thermal insulation and sound insulation. It is widely used in construction and aerospace by improving its uniformity and fiber diameter to further enhance its thermal and acoustic insulation properties. In this article, the purpose is further to create a smart manufacturing system using artificial neural network to provide analysis, judgment, and optimization for the manufacture of aerospace-grade ultra-fine glass fiber felt. When there were 11 neurons in the hidden layer, both the relative error Z values of the uniformity and the fiber diameter were the smallest, which were 0.0382 and 0.0073, respectively. So the structure 3−[11]1–2 with the back-propagation training algorithm was the most adaptive model, which was proved by comparing the mean relative error. In addition, after comparison with the measured data, the predicted and measured values are very similar and the error between them is small, so this structure has been confirmed to have a high accuracy. Finally, three-dimensional planes for the predicted uniformity and fiber diameter as a function of each process parameters are established. The predictive quality was pretty satisfactory, which can be applied to predict new data in the same knowledge domain.