Workpiece Dependency Exploration & Probabilistic Nonparametric Modelling of Vibration-Assisted Hybrid Micro-EDM Process

• Published in: Arabian journal for science and engineering (2011) 2022-12, Vol.47 (12), p.15331-15345
• Main Authors: Singh, S. K., Mali, H. S.
• Format: Article
• Language: English
• Subjects: Austenitic stainless steels; Engineering; Gaussian process; Humanities and Social Sciences; Hybridization; Machining; Mechanical properties; multidisciplinary; Nonparametric statistics; Probabilistic methods; Probability theory; Research Article-Mechanical Engineering; Science; Statistical analysis; Surface roughness; Tool wear; Vibration effects; Workpieces
• ISSN: 2193-567X; 1319-8025; 2191-4281
• DOI: 10.1007/s13369-022-06616-9

μ -EDM is one of the leading process used to fabricate micro-parts and features that find application in wide array of products such as MEMS devices and lab-on-chip systems. To mitigate limitations of μ -EDM, hybridization of the process is done using vibration assistance. However, ambiguity remains as to the extent of effect of hybridization on different materials. In this work, experimentations are performed to explore and assess whether the benefits, and disadvantages, of hybridization are dependent on type of material used. Three workpiece materials—AISI 316 SS, Co29Cr6Mo and BeCu, which have different thermal, electrical and mechanical properties are chosen. Microholes of varying depth are drilled at different vibration frequencies and discharge energies to assess the effect on machining time, tool wear and surface roughness. Thereafter process outputs are modelled using a nonparametric probabilistic method called Gaussian Process Regression and are validated on new design points. It is found that effect of vibration is highly material specific and benefit is mostly limited to reduction in machining time. Microscopic images of machined surface reveal extreme globule formation in AISI 316 SS and few micropores in Co29Cr6Mo. Gaussian Process Regression shows great capability to model seemingly stochastic output using limited training data.