NEURAL AND NERVOUS REFLEX NETWORKS: A NEW APPROACH TO INTEGRATED CONTROL
Alexandru M. Panait, Alexandru Mihail MOREGA · 2009
Reţelele neuronale sunt privite ca soluţie “exotica” la problemele nelineare si se prefera implementarea lor software pe masini cu arhitectura clasica, cum ar fi masinile Von Neumann sau echivalente, sau chiar pe controllere specializate, denumite “computere neuronale”. Aceasta lucrare reprezita o direcţie alternativa la cercetarea reţelelor neuronale artificiale, pentru ca propune o implementare hardware prin particularizarea un tip special de oscilatoare in inel inventate de Mark Tilden. Reţeaua neurala tip Tilden este adaptata pentru a comanda un tracker solar, necesitând foarte puţine componente si având un cost mult mai mic, pastrând fiabilitatea, robusteţea si flexibilitatea soluţiilor clasice ca parametri de baza. Ideea originala consta in adaptarea acestei clase de reţele pentru comanda directa de poziţionare a unui motor, si in conceptul de control distribuit in sensul asigurarii unei anumite autonomii efectorilor din sistemele complexe, folosind circuite cu raspuns imediat la modificarile unor parametri monitorizaţi. Neural networks have always been regarded as an exotic solution to nonlinear problems and often thought to be more effective as code implemented on a classical architecture machine, such as a Von Neumann machine or equivalent, or even given a controller class of their own, the “neural computer”. This work represents a breakpoint with that line of thought, as it proposes a hardware alternative to the neural network implementation, by particularizing a specific type of ring oscillator invented by Mark Tilden. The Tilden neural network is adapted to drive a solar tracker, requiring only a fraction of the component list and with a minimal cost, while keeping robustness and flexibility as core parameters. The original idea is adapting this class of neural networks to directly control the positioning of a motor, and describing a type of integrated control where distributed limited autonomy “slave” controllers also exist, enabling immediate response to changes in the monitored parameters.