Argimiro Resende Secchi

DAWRS: A Differential - Algebraic System Solver by the Waveform Relaxation Method

We investigate the concurrent solution of low-index ndifferential-algebraic equations (DAE’s) by the waveform nrelaxation (WR) method, an iterative method for system nintegration. We present our new simulation code, DAWRS n(Differential - Algebraic - Waveform Relaxation Solver), to nsolve DAE’s on parallel machines using the WR methods, and ndescribe new techniques to improve the convergence of nsuch methods. As experimental results, we demonstrate the achievable concurrent performance to solve DAE’s for na class of applications in chemical engineering.

Dynamic process simulation using a concurrent differential and algebraic solver

Abstract Rigorous modeling of dynamic chemical processes results in systems of differential and algebraic equations (DAEs) whose solution on sequential computers can be highly time consuming. A possible alternative for the solution of very large systems of DAEs is to use concurrent computation on multiprocessor computers, or multicomputers. The waveform relaxation (WR) method, an operator-splitting approach to the solution of such DAE systems, partitions the problem into several lower order systems (subsystems). These subsystems can then be solved using standard integration codes. This method results in algorithms with a highly parallelizable concurrent fraction and low sequential overhead, which are especially suitable for coarseand medium-grain MIMD distributed-memory machines. In this paper, we describe a new simulation code, DAWRS (Differential- Algebraic Waveform Relaxation Solver) to solve DAEs on parallel machines using the WR method. Using a rigorous model for distillation columns, we illustrate how this code can be efficiently applied to the simulation of dynamic chemical processes, and what the limiting factors (in terms of efficiency) turn out to be. A process network involving several distillation columns, with various configurations and dimensions, is given as an illustration of large scale dynamic simulation.

A Kriging-based approach for conjugating specific dynamic models into whole plant stationary simulations

Abstract Steady-state simulators are usually applied for design, techno-economic analysis and optimization of industrial processes. However, sometimes dynamic systems are important parts of the process, which cannot be disregarded. Coupling a dynamic model within a full-plant for steady-state simulation is a challenging task, whatever might be the simulator concept, either sequential or equation-oriented. An alternative to solve this problem is the use of surrogate models to substitute specific dynamic models, by taking the variable time as an extra input of the meta-model. This methodology was applied in an equation-oriented simulator (EMSO) by the use of Kriging meta-models. A case study involving the production of bioethanol from sugarcane was used to demonstrate the capability of this approach. A Kriging meta-model used to substitute the kinetic model of an enzymatic hydrolysis reactor was conjugated into the global plant simulation and an optimization problem was successfully solved.

Improving the Convergence of the SELEST Identifiability Procedure

SELEST is a procedure for identifiability of parameters in which selection and estimation steps are simultaneous, ensuring a well-conditioned estimation problem for a subset of identifiable parameters. Nevertheless, since SELEST is based on local sensitivity analysis, the identifiability criteria are dependent on the parameters initial values, requiring intensive parameters evaluation. In order to improve the convergence of the algorithm, we propose to update the values of the selected parameters and their sensitivity submatrix when re-ranking the remaining parameters. Therefore, the parameters estimations are performed using more appropriate values than the initial estimates. Two cases studies illustrate the performance of the proposed procedure: a hypothetical model, and an enzymatic hydrolysis model. Results demonstrate that the proposed modifications improved the performance of the algorithm, reducing the computational time significantly.