Projects / Programmes
Kemijska procesna sistemska tehnika (Slovene)
Code |
Science |
Field |
Subfield |
2.02.00 |
Engineering sciences and technologies |
Chemical engineering |
|
Code |
Science |
Field |
T350 |
Technological sciences |
Chemical technology and engineering |
T200 |
Technological sciences |
Thermal engineering, applied thermodynamics |
distillation columns, heat integration, process synthesis; batch operation, continuous operation, break-even capacity; MINLP, sequential/continuous optimization, superstructure; model prediction, neuro nets, phase equilibrium.
Researchers (7)
Organisations (1)
Abstract
The research comprised synthesis, design and optimization of processes as well as use of neuro nets in chemical engineering.
A new procedure for finding the best heat integrated distillation systems structure was developed. In the first approximation total utilities cost gives a good estimation. Distillation columns systems ranking with regard to the largest sum of the products between intercolumns heat flow rates and temperature differences between the condensers of one and the reboilers of the other column is equal to the ranking according to total annual costs.
Selection between batch and continuous processes has been testend on another example, dibutyl phtalate production. Previous results have been confirmed - under the limiting capacity batch processes using multipurpose equipment are more economic. Above it continuous ones are better. The limiting capacity depends on the process and the investment/annual cost ratio.
A combined sequential optimization method for process retrofits has been developed. It includes: superstructure generation using pinch analysis, NLP simplified MINLP model formulation, sequential direct search optimization and simultaneous optimization using GAMS algorithm. The optimization can be recycled several times.
Neuron nets are being developed to forecast the best phase equlilibrium models for different mixtures. Programs for models classifications, user''s data transformations, Kohonen map drawing and phase equilibrium model power were developed and the learning program modified - it is now 30 times faster than before. Learning and testing of nets will now be enabled in order to use them for model prediction.