Intelligent measurement systems based on neural networks.

AUTOR(ES)

Camargo, Laurizete dos Santos

DATA DE PUBLICAÇÃO

2000

RESUMO

Neural networks, control and systems theory and techniques are utilized in this work to improve the accuracy of measurement instruments. Contributions can be classified by subject under the major field they belong to. One contributions is the mathematical formulation of instruments based on system approach; it permits the global treatment of the measurements system without particularizing parts or having to specify the cause of the problems. Therefore, it guarantees that the advantages of systems approach are reached. Another contribution is the utilization of neural networks as estimators or as neurocontrollers. The neural estimators of measurement system functions are called emulator herein. Within this context the second method of Lyapunov is employed to study the stability and tracking of the system, resulting in a compensating measurement system with self adjustment. Another contribution is a basic procedure to building neural networks in a way that their capability to universal approximation to continuous functions is taken advantage of. Methodologies used in neural networks are reviewed, they are used to choosing the topology of the neural network and the number of hidden neurons. The innovation of this procedure is the utilization of polynomial interpolation theory, more specifically the Chebyshev theorem. It determines the size of the training set and indicates the elements of this training set to achieve the desired accuracy.

ASSUNTO(S)

funÃÃes de liapunov instrumentos de mediÃÃo inteligÃncia artificial redes neurais controle automÃtico interpolaÃÃo polinÃmios de chebyshev

Documentos Relacionados

• Aerodynamic coefficient prediction using neural networks.
• Sequential state generation by model neural networks.
• Chemical implementation and thermodynamics of collective neural networks.
• Analysis of E.coli promoter structures using neural networks.
• A more biologically plausible learning rule for neural networks.