Early Chatter Detection using MaxEnt and SPRT

In high-speed milling process, the early detection of chatter contains two aspects, one is the fast detection of the initial symptom of the chatter, the other is to recognize the slight chatter embedded in the noisy signal. This paper presents an online cumulative chatter detection method for high-speed milling process based on once-per-revolution sampling, maximum entropy (MaxEnt) principle and sequential probability ratio test (SPRT). The method allows for coping with these two aspects of early chatter detection. This method has less computational complexity and is independent of the cutting conditions. The procedure consists of four steps. First, the prior knowledge of early chatter is determined. Secondly, once-per-revolution sampling data is sampled from the vibration signal. Thirdly, the MaxEnt principle is used to estimate the MaxEnt of the once-per-revolution sampling data as a chatter indicator. Finally, the SPRT cumulates the information of the estimated MaxEnt, and then detect the early chatter by using the prior knowledge. The proposed strategy is applied to a high-speed milling process, and two simulation experiments allowed to assess the effectiveness of the early chatter detection method.