A mathematical neural network framework for real-time estimation of solid-phase pyrolysis parameters

Understanding solid fuel pyrolysis is essential for predicting material behavior in fires. This process is commonly analyzed through the mass loss rate (MLR) as a function of temperature, often using the Arrhenius equation. However, determining its kinetic parameters is complex, and numerous methods have been proposed. Recently, artificial intelligence techniques have emerged to improve parameter estimation, typically requiring each application thousands of simulations to optimise the fit to experimental data such as MLR curves. This study introduces an inverse modeling approach using a General Regression Neural Network (GRNN) trained on thousands of thermal decomposition simulations in a gasification apparatus. By varying thermokinetic properties, a comprehensive database is generated, reducing future computational costs. Users can then provide an MLR curve from bench-scale experiments—such as those in a gasification or Fire Propagation Apparatus (FPA)—to the prediction model. The GRNN framework delivers material characterization within seconds, offering a practical tool for obtaining input parameters for fire models like Fire Dynamics Simulator (FDS). Results show reduced uncertainty compared to the FDS Guide, demonstrating the potential of GRNN-based models to enhance the accuracy of solid-phase pyrolysis simulations in fire engineering.