The development of a new angioplasty balloon is a time consuming process. This thesis aims at reducing the amount of time and materials spent on the experimental stage of the development of new angioplasty balloons. This can be achieved by building a nonlinear neural network model of the balloon forming process and implementing an off-line cycle-to-cycle controller. The controller can learn from the previous experiments and provide better input parameters for improving the quality of the next balloons formed in the process. It is shown in the experimental test results that the neural network model can provide accurate estimates of the process outputs. The neural network model combined with a cycle-to-cycle control strategy has the potential to replace the trial-and-error approach to balloon development that is commonly applied today.

This study sought to investigate the application of artificial neural networks (ANN) and fuzzy inference systems (FIS) to variably saturated soil moisture (VSSM) redistribution modelling. An enhanced approach to such modelling, that lessens computation costs, facilitates input preparation, handles data uncertainty, and realistically simulates soil moisture redistribution, was our main objective. / An initial review of existing soil hydrology models provided greater insight into current modelling challenges and a general classification of the models. The application of AI techniques as alternative tools for soil hydrology modelling was explored. / A one-dimensional (1D) model based on ANN and FIS was developed. To estimate fluxes more accurately, multiple ANNs were trained and combined by way of an FIS. The main body of the model employed the ANN-FIS module to model soil moisture redistribution throughout the profile. When tested against the SWAP93 model, the ANN-FIS model gave a good match and maximum error of <8%; however, it did not show a notable computation cost shift. / The investigation proceeded with development of another ANN-based 1D modelling approach. This time, the soil profile or flow region, regardless of its depth, was divided into ten equal parts (compartments). The ANN was trained to estimate moisture patterns for a whole soil profile, from the previous day's soil moisture pattern and boundary conditions, and the current day's boundary conditions. The model was tested against SWAP93 where an average SCORE of 90.4 indicated a good match. The computation cost of the ANN-based model was about one-third that of SWAP93. / At this point the study sought to develop a 3D modelling approach. The ANN was trained to estimate the nodal soil moisture changes through time under the influence of six neighbouring nodes (in a 3D space, two on each axis). The model's accuracy was tested against the SWMS-3D model. An average SCORE of 91 and a 15-fold decrease in computation costs showed a quite acceptable performance. Results suggest that this approach is potentially capable of realistically modelling 3D VSSM redistribution with less computation time. / Finally, pros and cons of these ANN-based modelling approaches are compared and contrasted, and some recommendations on future work are given.

No description available.

No description available.

No description available.

No description available.

No description available.

No description available.

No description available.

No description available.