Deep Learning Seminar  /  September 10, 2020

Flowsheet Simulation and Optimization Supported by Machine Learning Methods


Reliability, feasibility and computationally efficiency of flowsheet  simulations are major prerequisites for online plant control and  optimization. However, convergence issues still disturb the automated  use of flowsheet simulators in industrially relevant applications: The  reasons for a non-convergent simulation run can either be numerical  issues or physical infeasibility of the operating conditions to be  simulated. This situation results in tedious, time-consuming  investigations for the process engineer, which at the end might even  compromise the simulation and its inarguable benefits themselves.  Rigorous models with included sustainability metrics are typically used  to ensure high quality of predictions and to obtain the objectives of  interest. In this contribution, machine learning methods are coupled to  CHEMASIM, the inhouse flowsheet simulator of BASF. Adaptive simulation  plans are generated, which according to the results are adjusted  sequentially. As a result, the border between feasible and infeasible  operating conditions (in the sense of a binary feasibility  classification) is obtained parametrically. Finally, the methodology is  applied in the multicriteria optimization of an exemplary process, where  economic and environmental targets are included.