Avaliação de controle neural a um processo de quatro tanques acoplados

AUTOR(ES)

Júlio César Peixoto de Oliveira

DATA DE PUBLICAÇÃO

2009

RESUMO

The main goal of this doctorate thesis is the evaluation of neural control on a four interconnected tanks process, in laboratory environment, where the controlled variable is the water level height of the fourth tank. The dynamics of this process is non linear, as the outflow from the tanks depends on the square root of corresponding water height. This type of process is quite common in industrial applications as, for example, chemical and petrochemical plants. Because the process is non linear, the performance of conventional control techniques depends strongly on the operation points, thus, on parameters adjustment for each operation point. Many different controllers have been implemented aiming the evaluation, research, validation and improvement of artificial neural controllers. The characteristic behavior of this process does not allow satisfactory performance on a wide operation range with the use of conventional controllers, what justifies the use of neural controllers. For the design, implementation and simulation of the neural controller on the process, a general purpose commercial computer had been used to run a Matlab/Simulink software environment. To implement the neural controller, electronic interface modules have been used. The adopted methodology was: identification, training, simulation and control for the proposed operation range using data captured from the plant. The NARMA-L2 structure (Nonlinear Autoregressive-Moving Average-Norm-L2) was initially used to model the plant, while the RNA plant model was used subsequently to calculate the control law. This algorithm, proposed by Narendra, K.S. and Mukhopadhayay, 1997, transforms, in the ideal case, the non linear system into a linear system through the addictive and multiplicative cancellation of non linearities. The advantage of neural networks over conventional control engineering is that it neural networks learn from the process data. Many simulations and experiments with neural controllers have been carried out on the process. Experimental results of neural control and comparative tables are presented throughout this thesis. The viability of neural networks with fixed weights for the control of non linear plants is demonstrated. It should be remarked that, for this process, there are cases for which the neural controller could not be tuned to deliver satisfactory performance. The main contribution of this thesis is showing an evaluation of NARMA-L2 neural control on a complex fourth order non linear process. So far, an unseen evaluation.

ASSUNTO(S)

liquid level system narma-l2 sistemas não lineares controlador programável narma-l2 sistema de nível de líquidos artificial neural networks identification redes neurais artificiais identificação non linear systems process control engenharia eletrica programmable controller controle de processos

Documentos Relacionados

• Controle preditivo robusto baseado em desigualdades matriciais lineares aplicado a um sistema de tanques acoplados
• Estudo comparativo do desempenho de controladores robustos aplicados a um sistema de tanques acoplados.
• Avaliação do desempenho operacional de um convertedor LD através do desenvolvimento do processo de sopro com lança de quatro furos
• Controle inteligente da qualidade de água para tanques
• Uma proposta de controle neural adaptativo para o posicionamento de um braço mecânico