Simulation-based optimal design of experiments for productivity improvement in biodiesel production

MARTIN F. LUNA; ERNESTO C. MARTINEZ

Buenos Aires

Workshop; Workshop on Applied Modelling and Simulation (WAMS); 2013

A significant source of uncertainty in biodieselproduction is the variability of feed composition sincethe percentage and type of triglycerides variesconsiderably across different raw materials. Also, due tothe complexity of both transesterification andsaponification kinetics, first-principles models ofbiodiesel production typically have built-in errors(structural and parametric uncertainty) which give riseto the need for obtaining relevant data throughexperimental design in modeling for optimization. Forincreasingly improving productivity in biodieselproduction, simulation-based design of optimallyinformative experiments is proposed. The presentapproach integrates tendency models with Bayesianactive learning to balance knowledge exploitation withsafe exploration in the steep direction of improvement.Parameter distributions are re-estimated using data fromexperiments and bootstrapping. Results obtainedhighlight that Bayesian optimal design of experimentsusing a probabilistic tendency model is effective inachieving the maximum yield in biodiesel productioneven though significant uncertainty in feed compositionand modeling errors are present.