Milling Tool Condition Monitoring using Vibration Signals and Histogram Features through Machine Learning: A Comparison of Naive Bayes and Bayes Net Algorithms

Abstract

In material removal process, cutting tool plays a vital role in producing the required surface finish. The continuous usage of cutting tool makes it to worn out. This even affects the surface finish, when the change tool geometry exceeds the threshold limit. Hence, optimal tool change is necessary and can be performed only through continuously monitoring the condition of the tool. Face milling is one of the prominent metal cutting process employs multi-point cutter and it is taken for the study. In this work, four conditions of the face milling tool namely tool in defect free condition or new tool, tool with flank wear, tool with flaking fault and tool with broken tip are considered. The continuous monitoring of the cutting tool while machining is carried out through vibration signal, since it correlates well with the tool conditions. The histogram plot is computed from the amplitude of the raw vibration signal. Then the machine learning algorithms such as Naive Bayes algorithm Bayes Net algorithm are utilized to differentiate the different conditions of the cutting tool with the input of histogram as feature set. The results show that both the algorithms classify the tool conditions well, where the Bayes net algorithm outperforms the Naive Bayes classifier with the classification accuracy of 91.75% with less error.