Multilayer perceptron network and Vyazovkin method applied to the non-isothermal kinetic study of the interaction between lumefantrine and molecularly imprinted polymer

Artificial Neural Networks for Pyrolysis, Thermal Analysis, and Thermokinetic Studies: The Status Quo

“…It should be mentioned that ANN does not always yield the results that agree well with the isoconversional ones. These two approaches have been compared as applied to the kinetics in the system lumefantrine/molecularly imprinted polymer [ 101 ]. The isoconversional method has demonstrated that in the range of conversions from 0.05 to 0.90, the activation energy rises from roughly 80 to 150 kJ mol −1 .…”

“…This is a sure sign of the kinetics being complex, i.e., to involve at least two steps with significantly different activation energies [ 102 ]. Yet, the ANN estimates yield the activation energy whose values fluctuate in the range of approximately 160–200 kJ mol −1 depending on the heating rate used in analysis [ 101 ]. Of course, one cannot expect ANNs to reveal the multi-step nature of the process kinetic as long as the networks are trained on single-step kinetics.…”

“…The general idea is to utilize the neurons in the form of reaction model functions. The authors apply a similar approach for the description of isothermal processes of widely differing materials, i.e., rhodium (II) acetate [ 17 , 18 ], efavirenz and lamivudine [ 93 ], thalidomide [ 103 ], polyurethane [ 104 ], hybrid of carbon nanotubes with poly(3-hexylthiophene) [ 105 ], composite Li 4 SiO 4 /MgO [ 106 ], calcium trihydrate/furosemide system [ 107 ], and lumefantrine/molecularly imprinted polymer [ 101 ]. To show the basics of this method, we will follow the first paper [ 18 ], where the ANN with a single neuron in the hidden layer is considered ( Figure 14 ).…”