Abstract
Electrical discharge machining has been employed for cutting special alloys and hardened steel for some decades; however, the process has still been treated as an empirical art rather than a technical skill because of the complex physical phenomena are incompletely understood during the process. With the advent of neural networks for modeling manufacturing process, the applications of neural networks on modeling of electrical discharge machining could be thoroughly studied and evaluated for better understanding the process; and, as a further step, the results could lead to the improvement of the process efficiency. In this dissertation, complete review on the published literatures has been conducted for including the pertinent process parameters of the process. Then, Design of Experiment has been established for screening the most important parameters in order to compare the effectiveness of various models, namely a semi-empirical model and seven types of neural networks models. The semi-empirical model has been based upon fundamental laws of physics together with dimensional analysis for pure electrode and work materials. Whereas, the neural networks have been completely accounted for the through study on the experimental data given by the design of experiment procedures. The erosions of both tool and work-piece and the surface roughness of the work-piece materials have been measured and analyzed for the purpose of testing the models. With the appropriate training process, the predictions from both the empirical model and the various neural networks models have been compared to the checking experimental results. Evidently, the neural networks have shown better agreement than the empirical model in this study. This could lead to the successful applications of modeling of electrical discharge machining on the shop floor in the near future.