| Computer hardware and software are tools of the trade for today's engineer. Neural networks are such programs. Their practicality allows chemical engineers to model an entire plant or a single unit, such as a distillation column. Using neural networks, it is not necessary to include specific dynamics of the plant. The neural network employed in this thesis was NeuroShell. It was used in an attempt to discover the robustness of trained model which contained gross errors of ten and thirty percent and a model with noise or disturbances plus some gross errors. There were sixteen defining characteristics and only one classifying characteristic. From the results, one can conclude that the introduction of gross errors into a network based on first principles does not deter the convergence of the network. |
