Parallel computing in model-based process engineering

Abstract: Chemical processes are usually expensive to run. To optimise a process, we need to know how the process behaves, and this is usually accomplished by studying the process experimentally. However, experiments are very expensive, particularly if an existing plant has to shut down normal operation to conduct an experiment. A better way is to build a model and simulate the process. In order to obtain the correct answer to an optimisation, the model needs to be accurately calibrated, where the model is tuned to fit experiment data. A simulation of a chemical process can take very long time and many simulations are needed for both the optimisation and calibration. As the computers are becoming faster and faster, more advanced modelling is possible. Parallel computing can be used to manage many time-consuming simulations in a reasonable amount of time. Parallel computing uses many computers, and simulates multiple jobs simultaneously. In this thesis, several cases are investigated that use methods that are heavy in terms of computation, such as multi-objective optimisation and parallel simulations and calibrations. The cases studied include parameter estimation in a polyethylene plant, parameter estimation in a combined cycle power plant and multi-objective optimisation of batch and continuously chromatography processes. The results presented could not have been reached in a reasonable time without the computer cluster available. Optimisations become very complex if all decision variables and objective functions of interest are to be included in the problem, and limits must usually be set in order to obtain the answer within a reasonable time. With more computational power, more advanced problems can be solved. Where discretisation is needed, such as in partial differential equations or trajectory optimisation, the solutions can be made more accurate by making the discretisation grid finer. The different kinds of parallelisations were classified into open loop, closed loop and hybrid loop vectorised computing, and all categories are needed for various kinds of projects. Computations were broken down to several levels of calculations, where parallelisation was possible between two levels. Two parallelisations were performed simultaneously for hybrid loop vectorised computing. Parallel computing is attracting greater interest as the number of processor cores increases, and the available methods for parallelisation are predicted to improve. The results obtained from our homebuilt cluster show the potential of multi-computer computations in model-based process engineering.