Optimization of CNC Lathe Operation by ANOVA and Validation using Gradient Descent
The following study discusses an investigation into the use of full factorial design methodology for parametric
study of turning operation response variable. ANOVA was used to identify the dependency of our response variable i.e.
Surface Roughness on input parameters i.e. Speed, Feed and Depth of Cut. 27 experimental runs were conducted to define
the ideal conditions of input parameter. To validate the dependency results from ANOVA we have applied a widely used
algorithm from Machine Learning - Gradient Descent. The algorithm iteratively models our response parameter as an output
of our input variables and updates the coefficients over a predefined number of iterations and learning rate. The coefficients
output from Gradient Descent validate the ones from ANOVA. MINITAB 17 software was used to perform the analysis for
ANOVA and the Gradient Descent algorithm was written in Python. Main effect plot is used to determine optimum value of
individual response. Interaction plots are used to determine whether there is an effect of one factor on the level of other
factors. Normality tests were conducted to validate assumptions involved in ANOVA. The optimum value of Surface
Roughness is 0.6 micron at speed 800 rpm, feed 40mm/min and depth of cut 1.0 mm.
Keywords - CNC Lathe, ANOVA, Composites, Silicon Carbide, Gradient Descent, Machine Learning