الفهرس يوجد فقط 14 صفحة متاحة للعرض العام
 |  | from | 124 |  |  |
| | | from | 124 | | |
المستخلص
ABSTRACT
Recently the term freeform surface (FFS) has drawn attention, specifically in Reverse Engineering (RE) field. The process of reproduction of a freeform surface is not an easy task. Not only due to its complicated nature but also because of insufficient measuring data points collected when using the Coordinate Measuring Machine (CMM). There are many different techniques used in surface approximation such as B-Spline technique, Bezier Curve, interpolation, and Artificial Neural Network (ANN). ANNs have a great acceptance due to their high self-adaptivity in estimating the missing points on the measured surface in case of 3-D or curve in case of 2-D. The most common kind that is used in surface estimation field is the Feed-Forward Back- Propagation (FFBP). BPNN is capable of producing good surface estimation in a reasonable training time. Unfortunately using ANNs requires a good level of experience while designing the network and selecting the parameters.
The work in this research is concerned with using the ANN to estimate the unknown space points between the measured points either in 2-D or 3-D measurements. The research proposes three enhancement techniques that can improve the performance of using the ANN technique. The first technique is helping in selecting the size of network to avoid under-fitting and over-fitting effects. The second technique aims at increasing the internal dimension of the network as well as increasing the ability for estimation. The third technique is the use of an addition type of neural network rather than BP to work as an error reduction technique. In these cases under considerations the Radial Basis Function (RBF) ANN is selected for this task.
Key words: Artificial Neural Network (ANN), Freeform Surface (FFS), Reverso Engineering (RE), Coordinate Measuring Machine (CMM), Radial Basis Function (RBF). Back Propagation (BP).