In this research, a data-driven approach to metamodeling of manufacturing/machining processes is developed. Instead of the conventionally used second-order polynomial regression metamodels, a non-predefined form-free approach is discussed. The highly adaptive metamodeling strategy, called symbolic regression, is carried out by using genetic programming. A central composite design based experimental dataset on electric discharge machining is used as the training and the testing data. Four different process parameters namely (voltage, pulse on time, pulse off time, and current) are used as the independent parameters to quantify three different responses (material removal rate, electrode wear rate, and surface roughness). The performance of the metamodels are evaluated by using various statistical metrics like R2, MAE, MSE. The performance of the metamodels on the training and testing data is found to be adequate for all the responses.
A data-driven approach to learning 3D shape