A study on tool condition monitoring and diagnosis of micro-grinding process based on feature extraction from force data

Abstract

In this article, a new algorithm for diagnosing tool conditions in micro-scale grinding process is proposed using features extracted from measured tangential grinding force data with the aid of wavelet packet decomposition and back-propagation neural network methods. The tangential grinding forces are measured in a series of micro-grinding experiments by varying depth of cut and feed rate, and those measured profiles are analyzed to define the tool conditions—sharp, middle and dull. From each tangential grinding force signal, 32 node energies are extracted by applying a wavelet packet decomposition method, and a total of 34 features including 32 node energies, depth of cut and feed rate are used to build the micro-grinding tool condition diagnosis model based on a back-propagation neural network approach. In this model, the grinding tool condition can be represented as a numerical confidence value. The experimental verification is conducted and it is demonstrated that the developed model is applicable for effectively diagnosing the micro-grinding tool conditions.