A new method for identifying the ball screw degradation level based on the multiple classifier system

Abstract The ball screw is one of the essential components of a machine tool and identifying its degradation level is therefore critical for the health management of the entire machine tool. This paper proposes a new method to automatically and reliably discriminate the defect level of the ball screw. In this method the multiple classifier system (MCS), rather than the single classifier system, is used for severity differentiation in order to enhance the classification accuracy. We adopt the dynamic classifier selection (DCS) strategy for the MCS and design a novel local class accuracy technique (N-LCA) to replace the conventional local class accuracy technique (LCA) for DCS, so that the performance of the DCS strategy can be significantly improved. Since the LCA selects the most suited classifier for a testing object by estimating each classifier’s competence in the sample’s neighborhood, how to define such a neighborhood has a direct influence on the performance of the LCA. In the N-LCA, the neighborhood components analysis algorithm is introduced to adaptively and accurately determine the neighborhood for better tackling the difficulty in a reliable local accuracy evaluation. Eventually, a new MCS is constructed with the N-LCA to better discriminate the ball screw degradation severity, and its effectiveness is verified by our experimental results.