Estimation of machining performance and machining characteristics using artificial neural network in wire electric discharge machine for titanium & P-20 materials

Wire electrical discharge machining offers a dynamic solution for shaping solid materials with intricate configurations. This specific machining process excels in accurately crafting components for hard materials and also for complex shapes material. The focus of the current investigation revolves around the machining of titanium grade-2 and P-20 tool steel materials employing the L16 orthogonal array. The study takes into account various process parameters, namely, current, bed speed, pulse on time, pulse off time, voltage and flush rate. Voltage and flush rate were kept constant during the machining operation, but the other four parameters are changed during the experiments. The electrode material employed for the present study is a 0.18-mm diameter molybdenum wire. A phenomena known as an acoustic emission occurs when localized sources within a solid quickly release stored elastic energy, creating momentary elastic waves that travel as a spherical wave. When subjected to appropriate processing and analysis, the waves which have been detected will be converted as an electric signal that can provide useful information on the source of energy releases. In order to determine the main influences on machining characteristics, such as electrode wear and surface roughness, and machining performance, such as acoustic emission signals optimization of process parameters was first carried out. Using the optimized parameters, simple functional relationship charts between the parameters were built to acquire data on acoustic emission signals, electrode wear, and surface roughness. However, this straightforward analysis does not offer comprehensive information about the material, electrode, and signals status. Consequently, a more advanced method of analysis, namely artificial neural network was employed for predicting (estimating) the experimental values. At 70% of the training data, compared to 50% and 60% of the training data, the correlation between the predicted values of the artificial neural network and the experimental data was found to be stronger.