Support Vector Regression for Determination of Minimum Zone

Abstract

Probe-type Coordinate Measuring Machines (CMMs) rely on the measurement of several discrete points to capture the geometry of part features. The sampled points are then fit to verify a specified geometry. The most widely used fitting method, the least squares fit (LSQ), occasionally overestimates the tolerance zone. This could lead to the economical disadvantage of rejecting some good parts and the statistical disadvantage of normal (Gaussian) distribution assumption. Support vector machines (SVMs) represent a relatively new revolutionary approach for determining the approximating function in regression problems. Its upside is that the normal distribution assumption is not required. In this research, support vector regression (SVR), a new data fitting procedure, is introduced as an accurate method for finding the minimum zone straightness and flatness tolerances. Numerical tests are conducted with previously published data and the results are found to be comparable to the published results, illustrating its potential for application in precision data analysis such as used in minimum zone estimation.