Modeling, optimization and assessment of surface topography in EDM by advanced methods

Abstract

Electric discharge machining (EDM) is one of the most versatile nontraditional machining processes in modern industries. Though application of EDM covers a wide range of industrial demand still complex mechanism of material removal restricts its maneuverability towards precision manufacturing. Presence of complex thermo-electrical phenomena, material inhomogeneity, transient behavior of dielectric fluid results complicated surface integrity. In the present study, a novel attempt is made towards the development of virtual environment of EDM process through modeling, pseudo Pareto optimization, inverse solution method and the understanding of randomness and chaos in surface topography. Experiments are conducted with different levels of three most significant machining control parameters namely current, pulse on time and pulse off time. For each experimental run, material removal rate, different surface roughness parameters are measured and scanning electron micrographs are taken. From the obtained results, some irregularities in the behavior of process outcomes are observed. To encompass the irregular fluctuations, support vector machine (SVM) regression is employed for model building purpose. During model building of process outcomes internal structural parameters of SVM regression are tuned using two evolutionary metaheuristic optimization methods namely particles swarm optimization (PSO) and teaching learning based optimization (TLBO). Generalized modifications namely population based termination criterion, weight combining method are improvised for better performance of optimization techniques. It is observed that modified TLBO is computationally cost efficient than modified PSO. Modified TLBO is further employed for developing a unique learning system for two process outcomes. Here combined rank method is proposed for efficient handling of multiple objective functions without affecting their individual impacts. Further, the representative SVM regression based learning systems for each of the process outcomes are used for optimization of EDM process. Pseudo Parent front is developed and a relation between optimum achievable combination of process outcomes is estimated. An inverse solution methodology is proposed to get the setting of machining control parameters in EDM machine to meet the customer demand based requirement. Moreover, an attempt is made to correlate the surface generation process with the characteristics of surface topography. Sequence of profile heights measured on each of the machined surfaces is considered as representative time series of that corresponding machined surface. Contributive effects of randomness and periodicity on surface topography is assessed through the formation of autocorrelation function. Predominance of randomness is observed through the evaluation of a non-dimensional index called as PR ratio. Also, presence of chaos in surface topography is checked. Saturation of correlation exponents measured on phase space, reconstructed from representative time series, indicates the presence of chaos. Non-integer value of correlation dimension suggests the fractal nature of machined surface. A different test directly from time series without phase space reconstruction, that is 0-1 test, is performed and presence of chaos for all machined surfaces is substantiated. However, the proposed methodology for model building of process outcomes, pseudo Pareto front development, inverse solution method to get settings of machining control parameters to meet specific need based requirement, evaluation of contributive effects of randomness and periodicity in surface topography and investigation of the presence of chaos could be implemented to any such process in a generalized way. KEYWORDS: Electric discharge machining; Support vector machine regression; Particle swarm optimization; Teaching learning based optimization; Autocorrelation function; Randomness; Chaos