Inspecting Process-Diameter Relationships of Forcespun PVP Ultrafine Fibers via RSM and ANN-PCA Approaches.

In this study, polyvinylpyrrolidone ultrafine fibers were produced by the Forcespinning process. As the functional performance of forcespun fibers is strongly dependent on their diameter, response surface methodology and artificial neural network combined with principal component analysis were applied, for the first time in this study, to predict the relationships between fiber diameter and a set of combined chemical, environmental and spinning parameters. With this aim in view, 40 sets of Forcespinning trials were performed by combining six main variables, namely: solution concentration, syringe orifice size, relative humidity rate, ambient temperature, rotational speed and collection distance. The collected samples appeared as a random network of smooth fibers with a mean fiber diameter ranging from 461 nm to 8.978 μm. Interestingly, the optimized neural network consisting of six input variables, five hidden layers with ten neurons and one output layer exhibited the best prediction accuracy, thereby providing a remarkable Mean Squared Error of 0.0498 and a correlation coefficient of 0.9954, between observed and predicted fiber diameter. Additionally, 3D plots of ANN computed data, allowed pointing out mutual interactions between effects of chemical, spinning and environmental parameters, in relation to fiber diameter. Furthermore, principal component analysis conducted on ANN computed data, enabled classifying the contribution of each parameter as follows: solution concentration, followed by syringe orifice size, relative humidity rate, ambient temperature, rotational speed and collection distance. These findings demonstrate, for the first time, the key role of environmental parameters in the Forcespinning process. [ABSTRACT FROM AUTHOR]