Analysis and Optimization of Complex Flowsheet Simulations with Parallel Computing and Machine Learning

Convergence problems are common when dealing with complex flowsheet simulations. These usually do not result from shortcomings of the numerical algorithms, but from an unfavorable or even impossible setting of the degrees of freedom for the underlying system of equations.

Due to the highly non-linear relationships of the variables of state, users of simulation software often cannot identify which design variables in which range or combination lead to unsolvable systems of equations. This leads to time-intensive trial-and-error simulation runs in practice.

Multivariate sensitivity studies (MSS) can be used to systematically and automatically analyze complex flowsheets in the multidimensional space of the design variables. Parallelization of the calculation of flowsheets drastically reduces the time required.

Machine learning methods can be applied, for instance, to identify and exclude areas where the flow diagram does not converge before the calculation. This artificial intelligence is used to control optimization algorithms for the evaluation of complex flow diagram simulations.

This approach is demonstrated using two examples from the field of distillation. Figure 1 shows the results space of an MSS for a rectification column for the separation of ethanol and water.