Simulação e otimização de reator de formaldeido, processo prata, usando tecnicas de inteligencia artificial / Simulation and optimization of silver formaldehyde reactor, using artificial intelligence techniques

AUTOR(ES)
DATA DE PUBLICAÇÃO

2007

RESUMO

The focus of this Thesis is the simulation and optimization of a silver catalyzed fixed-bed reactor for the oxidation of methanol to formaldehyde, using artificial intelligence techniques (genetic algorithm and artificial neural networks). Formaldehyde is a key chemical intermediate used mainly in the production of adhesives and resins for several market segments. Its production by the ?Silver? process is behind new plants operated with the modern ?Formox? process, catalyzed by ironmolybdenum oxide, as far as performance and optimization tools are concerned. A few kinetic models are found in literature for the Silver process but they are inadequate to simulate the reactor at industrial conditions. Lack of good quality available kinetic data makes difficult the development of rate equations and new data is not easily obtained due to constraints to run experiments at relevant industrial conditions. A hybrid simulator was developed for the formaldehyde reactor, using a deterministic model based on differential mass balance equation over the fixed-bed, and a neural network to model the reaction kinetics, implemented as a sub-routine in the simulator. As kinetic data were not available for the neural network fit (training), this operation was performed through three approaches, using an association of genetic algorithms and classical training methods, employing process information (conversion and selectivity), widely available in literature and easily measured by industrial plants. After training the neural network through the three presented approaches, the simulator was able to properly estimate the reactor performance, which was validated by comparison with literature experimental data and industrial data. Many simulations were run in order to clarify the influences of important operational variables (temperature, air flowrate, methanol flowrate, water flowrate and residence time) on the reactor performance, presenting the results in graphic format. The discussion was carried out based on the reaction mechanism and literature work, adding value to the industrial and scientific community. The reactor optimization, using the validated simulator, was made by two methods: genetic algorithm and SQP. SQP was used as a reference, consisting in a classical optimization method, gradient-based. It was demonstrated that SQP may reach a local optimum, far from the global one, depending on the initial estimate. The genetic algorithm, differently, always converged to the global optimum, not depending on initial estimates. The best approach, however, was the association of both methods in order to obtain best precision with shorter computing times. The artificial intelligence techniques (neural network and genetic algorithm) were perfectly employed here together with deterministic equations, producing hybrid models. The kinetic modeling of the chemical reaction, as well as the reactor modeling, was successful, validated in several situations. Many cases were studied, indicating the optimized conditions for each one, whose conclusions will allow Silver plants to define the best operational policies to minimize costs with raw material consumption

ASSUNTO(S)

algoritmos geneticos otimização silver redes neurais (computação) prata formaldehyde formaldeido optimization genetic algorithms artificial neural networks artificial intelligence inteligencia artificial

ACESSO AO ARTIGO

Documentos Relacionados