Abstract
Abstract
The quality of steel balls has a substantial influence on bearing performance. The conventional way of measuring steel balls is by examining randomly selected specimens owing to the lack of efficient test methods. However, this cannot guarantee that every individual ball meets a certain standard. The traditional measurement is not acceptable in those fields where stability and reliability are highly demanded. An automatic method to measure a steel ball's quality to make exhaustive examination feasible and efficient was investigated. This paper describes the prediction model of a steel ball's vibration level based on image texture features by regression analysis method. The model establishes the relationship between the surface microscopic image of a steel ball and its vibration level, which is the major criterion to judge the quality of a steel ball. Experimental results have demonstrated that the prediction model approximates the real vibration level very closely. Several fitting models have been tested and a non-linear quadratic prediction model gives rise to the highest fitting precision.