Thermal degradation kinetics, mechanism, thermodynamics, shape memory properties and artificial neural network application study of polycaprolactone (PCL)/polyvinyl chloride (PVC) blends

Abstract

The thermal degradation dynamics of different composition of polycaprolactone (PCL) /polyvinyl chloride (PVC) blends was studied in detail. The thermal degradation kinetics of PCL/PVC blends in different compositions (coded as PCL/PVC:70/30, P1; PCL/PVC:50/50, P2; PCL/PVC:30/70, P3) at four heating rates (10, 15, 20 and 25 degrees C/min) were investigated. Flynn-Wall-Qzawa, Kissinger and Tang isoconversional methods, which do not depend on the reaction order, were used to calculate the activation energy (Ea) of PCL/PVC blends and Coats-Redfern method which is an non-isoconversional model was used to characterization the solid-state reaction mechanisms. For all blends, it was observed that the Ea values obtained with isoconversional models and the values obtained with non-isoconversional models were very near to each other and the average values were, for P1, Ea = 113.45 kJ/mol; for P2, Ea = 95.28 kJ/mol, and for P3, Ea: 87.79 kJ/mol. In addition, the D3, F2, A2 mechanisms were suggested for the P1, P2 and P3 blends, respectively. DSC results showed the transition attributed to the melting point for P1 and glass transition temperature for P2 and P3. DSC results revealed that blends with a high PCL ratio have a crystal structure, while blends with a higher PVC ratio have an amorphous structure. Shape memory recovery test results for the blends revealed that the PCL/PVC (70/30) (P1) blend exibited great strain recovery. Furthermore, in this study, decomposition temperature, heating rate and percentage of PVC in blends were taken as input data to improved an efficient artificial neural network (ANN) model, and the percentage of weight remaining during degradation of PCL/PVC mixtures was taken as output data. The 3-10-10-1 topology with LOGSIG-TANSIG transfer function and feedforward backpropagation was used as the artificial neural network model. Then, an effective ANN model was developed by taking the degradation temperature, heating rate, %pvc and %weight left data as input data, and calculated activation energy values as output data. 4-10-10-1 topology with LOGSIG-LOGSIG transfer function and feedforward backpropagation was used as ANN model.