Predicting the Temperature-Dependent Tensile properties of Polyphenylsulfone using a Machine learning approach

After 200 years of their birth, synthetic polymers are present in over 60 primary forms. Many are largely known and spread, others less common and investigated, such as Polyphenylsulfone (PPSU) which, on the contrary, offers multiple applications. Among them, in particular, many concerns the use of the PPSU at (cold or hot) temperature. But experiments with its temperature-dependent characterization are very few and limited to specific formulations. Machine learning algorithms can fill the gap providing accurate predictions. Specifically, an unsupervised classification was here used for clustering material data with the scope to recognize patterns between PPSU up to the selection of similarities respect to a polymer as reference. Then, a supervised regression was used to predict temperature-dependent tensile properties. A high level of accuracy was achieved, up to 99% in terms of coefficient of determination. This is probably the first time that data regarding the mechanical behavior of PPSU were derived from an approach based on artificial intelligence and machine learning.