Online damage detection of cutting tools using Dirichlet process mixture models

The ability to monitor and predict tool deterioration during machining is an important goal because the state of wear has a significant influence on the surface quality of machined components. To build up a comprehensive condition monitoring system for diagnosis and prognosis, however, extensive measurements and knowledge of tool wear is required. Collecting labelled datasets that include damage information for this purpose can be expensive and time consuming.

This paper suggests an unsupervised clustering approach using Dirichlet process mixture models to detect the change in characteristics of a cutting process online for diagnosis. As well as providing a useful monitoring tool, this approach has the potential to reduce the need for exhaustive wear measurements associated required for prognosis. The model is well suited to the erratic and unpredictable nature of tool wear progression, as the number of clusters required to determine the possible damage states are not set a-priori. Consequently, this method is equipped to handle variations across homogeneous and heterogeneous groups of tool material compositions.

The proposed approach is demonstrated here as a method to reduce the time required for trials for wear characterisation of new tools. In the example shown, the results indicate that the approach would result in around a 30% reduction of test times (on average) during outer diameter turning of case hardened steel, across 10 Polycrystalline cubic Boron Nitride tools from two different material compositions.