Closed‐loop poly(lactic acid) reprocessing: an artificial neural network for tensile strength decrease classification

Plastic waste management is an issue that requires innovative solutions. Currently, most plastic products go through a linear lifecycle (take, make, use and dispose of). To enable the alternative, closed‐loop recycling, it is paramount that manufacturers can know, beforehand, the expected properties of parts produced from a given recycled polymer. Therefore, solutions for product properties’ prediction are needed that use preexisting or in‐process collected data. Artificial neural networks (ANNs) are one of the most versatile options, as they allow modeling both linear and nonlinear data patterns. In this work, a database was compiled from the available literature reporting poly(lactic acid) (PLA) mechanical properties over multiple processing cycles. Information about number of previous processing cycles and polymer grade was used as input of an ANN classifier that predicts if the polymer's tensile strength has decreased below a given relative threshold. Strategies were used to improve the reliability of performance evaluation given the small size of the acquired dataset, as well as to account for class imbalance, namely, leave‐one‐out cross‐validation and synthetic minority oversampling technique, respectively. From a model with a single seven‐node hidden layer, it was possible to achieve average 91% and 90% classification accuracy and F1 score, respectively. The results are promising and indicate that, in the right context, good predictions will be achievable. This work also highlights the need for creating representative and comprehensive databases regarding the effect of closed‐loop processing on plastics properties, so better predictions can be made using directly measured material properties.