IMPROVING PHYSICS-BASED ENGINEERING FRAMEWORK FOR COMMODITY CLASSIFICATION

Abstract

This thesis evaluates and improves a zone-based fire model for predicting Critical Delivered Flux (CDF), a key metric for sprinkler-based fire protection in warehouse environments. Originally validated for Class 2 commodity, the model is extended to four other representative commodities: Class 3, standard plastic, plastic pallet and cartoned meat tray. The model represents rack storage as vertically stacked fuel zones and simulates fire growth and water transport using energy and mass conservation. A two-level hierarchical optimization framework is introduced to improve parameter estimation, classifying parameters as either commodity-specific or test-specific and optimizing them at different levels. This separation enhances parameter consistency and improves the model’s adaptability across different commodity types. The enhanced model is validated against experimental data for 5-tier storage configurations and demonstrates good agreement between predicted and experimentally derived CDF values for most commodities except plastic pallets. Two further refinements are incorporated: (1) commodity-specific tier-to-tier vertical ignition delays are applied to all commodity types to better represent observed vertical fire spread behavior; and (2) a suppression-driven ignition cut-off is applied specifically to the plastic pallet commodity, preventing upper-tier ignition after water application as observed from the experiments. The enhanced model is validated against experimental data for 5-tier rack storage and produces CDF predictions within acceptable accuracy limits for all commodity types. Overall, the proposed framework offers a scalable, physically grounded approach for commodity classification and fire protection design using only two lower-tier tests. In addition to validating and improving the zone-based model and its implementation, this study examines methods for identifying water application time from test data. Accurate identification is critical, as it defines the transition between free-burn and suppression and governs parameter calibration from lower-tier experiments, with errors propagating into predicted CDF values for high-tier configurations. When direct measurements are unavailable, visual observation of the total HRR curve is the most reliable approach. A sensitivity analysis is also conducted to evaluate the impact of the final test duration, showing that reducing it has minimal effect on both experimentally derived and model-predicted CDF values.