Application of machine learning to predict the properties of wood- composite made from PET, HDPE, and PP fibres

Abstract

Plastic composites provide an eco-friendly substitute for conventional construction materials. Indeed, recycling waste plastic represents a progressive approach to waste management with the aim of mitigating the growing issue of pollution in urban environments. Our research aims to review the physical properties, including water absorption (WA) and thickness swelling (TS), and mechanical properties, such as the internal bond (IB), the modulus of rupture (MOR), and the modulus of elasticity (MOE), of the latest findings made of wood panels combined with plastic. We are focusing on three types of plastic, namely polyethylene terephthalate (PET), polypropylene (PP), and high-density polyethylene (HDPE). In addition, we employed machine learning (ML) algorithms, including the hierarchical clustering dendrogram, the Pearson correlation coefficient, the support vector regression, the random forest (RF), and the decision tree (DT) for prediction analysis. For instance, the results indicate that combining HDPE with wood pulp fiber increases the MOR (42.45 MPa) and MOE (66.7 MPa), respectively. Furthermore, mixed plastics such as PET, HDPE, PP, and LDPE improve the dimensional stability by reducing the WA (0.32 %) and TS (0.18 %), respectively. In most cases, these results meet the minimum standard requirement for general-purpose boards, according with the American National Standard for Particleboard (ANSI/A208.1-1999), the European standard (EN 312), and Brazilian Association of Technical (ABNT NBR) standard. In addition, the dendrogram identifies three primary clusters with varying Euclidean distances, indicating the performance of wood-plastic panels for both physical and mechanical properties. Notably, the dimensional stability among panels is stronger than that of mechanical properties. The correlation matrix is important for selecting an appropriate plastic. The SVR, RF, and DT algorithms make predictions by analyzing the properties of the panel. For instance, the DT algorithm shows that when WA is less than 25 %, the predicted value of TS is 0.24 %; in addition, when the value is between 25 % and 75 %, TS is equal to 7.92 %; also, when WA is greater than 75 %, TS is predicted to be at 13.7 %. This innovative method of utilizing ML and DL for prediction opens new possibilities for the use of plastic in panel production, as it allows for the selection of suitable materials and fabrication techniques to create a wood-plastic composite.