Accounting for the moisture state of polymeric materials when developing machine learning models

The paper provides the results of studying the dependence of elastic-strength properties of unfilled epoxy polymers on moisture content by example of 18 different compounds. It includes the analysis of possible effects associated with changes in free moisture content in the polymer matrix structure, including a change in nature of behavior under load from brittle to viscous with a multiple increase in relative deformations at rupture, as well as quasi-embrittlement, manifested in the elimination or reduction of forced highly elastic deformations on the deformation curve. In addition to the shape corresponding to the one close to linear dependence of the change in tensile strength and modulus of elasticity on moisture content with a maximum around W~0%, the study revealed other forms of interrelation of parameters under consideration: with a local maximum of values in the “optimal moisture content” area differing from W~0%; with “plateau” sections around both extreme humidity conditions. The similarity of the effects occurring around moisture content of W~0% for epoxy polymer samples both in the control state and after prolonged climatic aging is shown. A hypothesis was formulated regarding the existence of a pattern common to epoxy polymers of change in the nature of the dependence of mechanical strength on moisture content during field climatic aging. Based on a joint analysis of the dependence curves of elastic-strength parameters on moisture content, the most representative epoxy polymer compounds for field studies were selected to compile training datasets for a machine learning model predicting changes in the elastic-strength properties of polymer materials exposed to environmental factors.