Adaptive-VDHMM for prognostics in tool condition monitoring

Among techniques used in condition monitoring, those for prognostics are the most challenging. This paper presents a Hidden Markov Model (HMM) based approach for prognostics in TCM. A HMM model usually employs a typical working condition for establishing and verifying the model. However, in tool condition monitoring (TCM), the cutting tool encounters a range of cutting conditions. It is not economical to establish a HMM for every cutting condition. Therefore, an adaptive-Variable Duration Hidden Markov Model (VDHMM) is proposed whereby the training information is adapted to a target test under different cutting conditions to those for establishing the initial model. It is found that with an appropriately selected feature set and state number, the proposed algorithm can significantly reduce the mean absolute percentage error (MAPE).