Multi-objective optimization of explosive waste treatment process considering environment via Bayesian active learning

A fluidized bed is a next-generation explosive waste treatment reactor that is safer and emits less pollutants (e.g., NOx) than a rotary kiln. When a fluidized bed reactor is used to treat explosive waste, the design and operating conditions significantly impact the emission of pollutants. It is possible to reduce the pollutants to below the regulation level (90 ppm) through finding ideal design and operating conditions. However, there are many practical challenges, such as cost limitations. In addition, owing to the characteristics of explosive waste, designing and optimizing the process through real experiments without any guidelines is dangerous. Therefore, a computational fluid dynamics (CFD) simulation was performed to obtain high-accuracy data on the internal phenomena of the reactor first. In this situation, since a lot of variables and combinations should be considered, it is obvious to takes very long time for finding optimal point by only using CFD simulation. Thus, based on the simulation data, efficient search space exploration was performed using multi-objective Bayesian optimization and several promising points constituting the Pareto front were derived to find optimal conditions. As a result, six optimum points of operating and design conditions were obtained considering process cost and nitrogen oxide emissions simultaneously. The six Pareto solutions through above approach reduced 47.5% of NOx emission and 10.5% of cost compared to the previous studies. In addition, it is meaningful that this study could reduce optimization time even though the design conditions of explosive waste treatment process were considered.