Researchers have developed a new system that uses a combination of machine learning and design experiments to provide sustainable and efficient electrospinning processes for the production of polymeric nanofibres. While electrospinning technology has received widespread attention, little research has been conducted on simulation studies, limiting the knowledge of the factors affecting the diameter variations of nanofibres. However, the system developed by the researchers used response surface methodology (RSM) to estimate the diameter of the electrospun nanofibre membrane and to identify the optimal values for various variables. The system included locally weighted kernel partial least squares regression (LW-KPLSR) to predict nanofibre membrane diameter. The accuracy of the system was evaluated based on measurements of mean absolute error (MAE) and root mean square error (RMSE), as well as the coefficient of determination (R2).
The diameter of nanofibres plays a significant role in determining their functional properties, such as filtration, adsorption, and catalytic degradation. Several factors, including voltage, needle size, relative humidity, and concentration of the polymer solution, among others, influence the diameter variations of nanofibres. Traditional optimization methods can be time-consuming and expensive, and they may not always lead to optimal values for all variables. Researchers have thus turned to machine learning models, which are faster and more efficient, to optimize these processes. Partial least squares (PLS) modelling is commonly used, but it has limitations, particularly in its ability to account for the non-linear relations between the data. Therefore, researchers have proposed an adaptive version of locally weighted partial least squares (LW-PLS) that accounts for the fluctuations in nonlinearity.
Meanwhile, the sparse kernel feature characterization factors are used in the proposed LW-KPLSR method to establish the importance of each training sample. The LW-KPLSR model performs better than other models in forecasting nanofibre membrane diameter, with significantly lower MAE and RMSE values. The system thus combines machine learning prediction models with design experiments, creating a sustainable and effective electrospinning process.
The researchers’ work is particularly important given the growing interest in electrospinning technology and its potential biomedical, tissue engineering, environmental engineering, and mechanical engineering applications. It mitigates the challenges posed by traditional optimization methods and paves the way for precise predictions about the form and characteristics of nanofibres. Moreover, it could open new avenues for electrospinning technology applications.
