Implementation of Statistical Learning Methods to Develop Guidelines for the Design of PLA-Based Composites with High Tensile Strength Values

Statistical analysis employing regression trees is utilized, for the first time, for a PLA-based biocomposite system aiming to extract knowledge in order to guide researchers toward intelligent selection of experimental conditions for desired tensile strength values. For the construction of the database, experimental data on PLA-based composites from past publications was collected using online sources such as ScienceDirect, Elsevier, ACS, and Google. The final data set that was built using 26 papers (out of similar to 50 initially screened) published between 1999 and 2018 contained 135 experimental data points. The response variable was selected as tensile strength, and 23 features regarding composite synthesis were included involving manufacturing method and temperature for compounding and testing, molecular weight of the PLA used, chemical composition of the composite, and type of filler employed in synthesis. Unbiased cross validation error of the regression tree was found to be 12-17 MPa, with coefficient of determination for prediction (R-p(2)) value equal to 0.5-0.7 showing moderate to -high prediction accuracy. Results indicated the following: (i) The feature in top node of the tree is the method used for manufacturing/testing rather than the type of filler employed and the composition of the composite (i.e., PLA content). PLA-based composites manufactured by solvent casting and direct mixing displayed significantly lower tensile strength values whereas biocomposites manufactured through other techniques involving compression, injection, melt blending, hot pressing, and aqueous suspension result in relatively higher tensile strength values. (ii) The molecular weight of the PLA had a significant influence in predicting the final tensile strength of the composite. The composites manufacatured employing PLA with molecular weight in the range of 275-400 kDa displayed high tensile strength values. (iii) The temperature employed during test specimen preparation for tensile strength measurements (following compounding step) seemed to play a critical role in the determination of tensile strength, and an optimum test temperature to yield the highest tensile strength possibly exists for each manufacturing method.