A modified Mahalanobis-Taguchi System analysis for monitoring of ball screw health assessment

Abstract

The ball screw's health assessment is significant to keep accuracy and reliability of the motion axes in the CNC machine. Mahalanobis-Taguchi System (MTS) is considered to be an effective non-parametric approach to carry out the health assessment. In this paper, a Laplacian Mahalanobis-Taguchi system (referred as LMTS) analytical model is proposed to establish a nonlinear mapping relationship between the features of sensor information and the ball screw performance. In order to utilize the limited sensor data effectively, LMTS method is only performed on the speed and motor current signals which are available in CNC secondary-develop interface. Because of the complexity of processing high dimensionality nonlinear features, Laplacian Eigenmaps is utilized to reduce the feature data dimension before they were sent to Mahalanobis-Taguchi System as inputs. Compared with the classical dimension reduction methods, the intrinsic low dimensionality manifold by Laplacian Eigenmaps in Mahalanobis feature space characterizes the performance degradation more accurately and robustly. Among many ball screw assessment technologies, this LMTS assessment is a promising data driven based approach because of less influence in the machining process and few changes in the original structural design. The results show that LMTS monitoring may enable the practical application of online real-time assessment for ball screws.