Stereolithography Apparatus(SLA) can produce accurate Rapid Prototype models for a wide range of applications. These can include aids to design verification. A major barrier to overcome before SLA can be used more widely is an acceptable part accuracy. In SLA, it is very important to reduce the dimensional errors of a products to improve the part accuracy..
This thesis selects the optimal condition of the process parameters to minimize the dimensional errors of the parts built by SLA. Because the accuracy of a part depends upon process parameters to determine the laser power, laser scanning velocity, and layer thickness. Taguchi method and genetic algorithm(GA) are used as tools to select the optimal condition of process parameter. The objective using Taguchi method is to select the optimal condition of the process parameters to maximize the S/N(signal-to-noise) ratio and reduce the number of the process parameters. For a method to select the optimal condition with GA and NN, a NN is constructed in order to model the relationship between the process parameters of SLA and the dimensions of H-part, the standard part. The training data to learn NN are obtained by full factorial experiment of three process parameters pooled. These are the process parameters which have an important influence on the dimensional errors of H-part. This true proves by the contribution analysis of the process parameters using Taguchi method. Using NN as estimator which predicts the dimension of H-part for the condition of a process parameter, GA searches for the optimal set to maximize the fitness function which is defined as the reciprocal of sum of square of the dimensional errors of H-part. In a comparative study, the experimental results of optimal sets to be obtained by Taguchi method and GA proves to have little dimensional errors than that of the nominal condition. And a sensor system to discriminate the solidified parts from liquid resin is proposed as the fundamental research to monitor the process of SLA.
