Tool-wear on-line estimation using a Dirichlet process mixture model

A new method based on Dirichlet Process Mixture Model (DPMM) is proposed for tool wear monitoring and tool wear estimation. This method describes the toolwear process as a wear accumulation process. Thus, the current tool wear is estimated by continuously estimating the wear increments. Firstly, the features are extracted from the raw force signals, and DPMM is used to classify these features automatically without determining the number of states of wear increments. Then, Gibbs sampling method is applied to identify the parameters of DPMM, which constructs the relationship between force signal features and wear increments. Based on the mixture model and on-line force signals, the wear estimation can be achieved. A practical study demonstrates that the proposed method is capable of self-adaptively learning wear increment states and effectively estimating the continuous tool wear.