Data-driven identification of LDPE pyrolysis kinetics using SINDy

Kinetic models for plastic pyrolysis must be both predictive and interpretable to support reactor design and scale-up. In this work, we examine whether the Sparse Identification of Nonlinear Dynamics (SINDy) framework can recover known lumped kinetics for low-density polyethylene (LDPE) pyrolysis. A four-lump scheme is used for the condensed phase, comprising polymer (P), heavy wax (Hw), light wax (Lw), and oil (Oil), with gas (G) treated as a cumulative sink. These five species are connected by nine first-order reactions. Pyrolysis temperatures between 250 and 500 °C are considered, with simulations carried out for 150 min at each temperature. Continuous-time SINDy with a linear library is then applied independently at each temperature to identify the governing operator for the condensed species and reconstruct all nine rate constants. SINDy accurately recovers the temperature dependence of the dominant pathways associated with polymer cracking and light-wax secondary cracking, and the resulting model reproduces the reference pyrolysis trajectories across all temperatures. In contrast, reactions with very small true rates, such as direct polymer-to-oil, oil-to-gas, and heavy-wax-to-oil conversion, are not reliably identified, reflecting intrinsic identifiability limits. These results demonstrate that SINDy can use pyrolysis data to discover reaction kinetics, reliably capturing the major kinetic pathways while naturally deemphasizing dynamically irrelevant reactions.