Title: Reduction of detailed CH4/NH3/H2 combustion mechanism using ANN-based global sensitivity analysis Abstract: Developing reduced combustion kinetic mechanisms that can predict key diagnostic parameters of the combustion process (such as ignition delay, flame speed, and pollutant emission) is essential for integrating finite-rate chemistry processes in combustor and burner simulations of novel fuels. In this work, a mechanism reduction strategy using Sobol sensitivity-based Global Sensitivity Analysis (GSA) is applied to reduce a detailed kinetic mechanism (153 species, 1350 reactions) for a ternary blend of ammonia, methane, and hydrogen. The unique contribution involves development of Multi-Layer Perceptron (MLP) based ANN surrogate models that were trained using A-factor perturbation of the detailed mechanism rate coefficients. These ANN surrogates were used to evaluate the Sobol based GSA indices. Global Sensitivity indices were determined for both reaction targets (ANN-GSA) and species targets (ANN-STGSA) and were subsequently used for mechanism reduction based on Ignition Delay Times (IDT) or Flame Speed (FS) predictions from the detailed mechanism. Overall, the ANN-STGSA technique demonstrated superior mechanism reduction potential. IDT-based ANN-STGSA produced the smallest mechanism (48 species, 370 reactions) but could not accurately predict non-target outputs like flame speed or concentrations. FS-based ANN-STGSA yielded a slightly larger mechanism (59 species, 679 reactions) that accurately predicted FS, IDT, and across a wide range of conditions. The FS-based reduced mechanism was integrated into a flamelet-based CFD code and successfully simulated major characteristics of methane and methane-hydrogen turbulent jet flames. Thus, the current work shows the potential of the ANN-STGSA technique as a computationally efficient approach towards robust multi-target reduction of complex mechanisms.
Title: Smart optimization and sensitivity analysis of lignin depolymerization: Bayesian learning approach to process and catalyst design Abstract: Fossil fuels dominate global energy production but are major contributors to greenhouse gas emissions, driving climate change and depleting natural resources. The lignocellulosic biomass has emerged as a powerful alternative due to its potential to produce diverse biofuels and chemicals sustainably. An abundant biomass component, lignin holds immense promise as a source of value-added chemicals. This study pioneers the optimization of lignin depolymerization to maximize guaiacol yield for biofuel and chemicals production, leveraging Gaussian Process Regressor (GPR) based Bayesian Optimization (BO). By utilizing γ-alumina supported nickel (Ni) and ruthenium (Ru) bimetallic catalysts, this integrated computational and experimental framework achieves simultaneous optimization of reaction and catalyst composition parameters with minimal experimentation. Unlike the traditional Design of Experiments (DoE) approaches that demand extensive experimental runs, BO accelerates discovery by optimally placing the experiments with just 20 trials, reducing the experimental workload by 99 % compared to full factorial design and 72 % compared to the Taguchi method. This breakthrough provides a sustainable pathway to convert lignin into valued feedstock chemicals like guaiacols. In the present investigation, the maximum yield of guaiacol achieved was 79.08 % under the optimized conditions, time: 1.87 hr, temperature: 551 K, lignin to solvent ratio: 0.0178 (wt./vol.), isopropanol to water ratio (vol./vol): 1.395, catalyst to lignin ratio (wt./wt.): 0.093, Ru: 5.29 wt%, and total metal loading: 19.34 wt%. Gratifyingly, this study unearths the most influential features driving guaiacol yield by introducing the Global Sensitivity Analysis accessing the most optimal GPR surrogate from BO, offering deep insights into the optimization results. This innovative approach not only advances the lignin valorization process development but also contributes to the broader goal of sustainable energy and chemicals production.