Process-machine interactions and a multi-sensor fusion approach to predict surface roughness in cylindrical plunge grinding process
Conference Paper
Overview
Research
Identity
Additional Document Info
Other
View All
Overview
abstract
2018 The Author(s). Published by Elsevier B.V. Abrasive finishing processes, such as cylindrical plunge grinding are employed across broad industrial sectors for surface quality assurance. However, uncertainties in the workpiece and wheel states due to the highly dynamic nature of the system are known to adversely affect our ability to predict and assure surface quality. Recent advances in smart manufacturing and sensing technologies offer an opportunity to improve surface quality assurance. Central to the sensing scheme for this is the identification of the physical relationship connecting the measured vibration signals with the surface characteristics. While our experimental studies suggest that vibrations over 700-2000Hz are significant determinants of the surface finish (Ra) in the plunge grinding process, prior models do not capture the effect of high frequency signatures (> 500Hz). To understand the physical origin of these high-frequency signatures, a three degree of freedom lumped mass model of the cylindrical plunge grinding process dynamics was investigated in this paper. The model integrates the random distribution of the abrasive particles forming the wheel topography, the interactions between an elementary abrasive particle and the workpiece surface, and the regenerative relationship between the vibrations of the machine structure and the cutting forces. The model was verified on an industry-scale "next generation precision grinder (NGPG)" testbed at the Indian Institute of Technology, Chennai, India. The experiments were conducted at twelve different plunge rates and data was gathered synchronously from two accelerometers and a powercell integrated with the machine tool. The results suggest that the model accurately captures the changes in the high frequency characteristics with the process parameters. Subsequently, a data-driven random forest model was developed to predict Ra using the features extracted from the measured signals. The model was able to predict Ra over various test conditions to an accuracy of R2>90%. Subsequent statistical analysis also suggests that the vibration frequency component in the 700-2000 Hz range contributes significantly to variations in surface finish Ra.
