Multiclassification of tool wear with support vector machine by manufacturing loss consideration

Tool wear is a dynamic process, as a tool progresses from sharp to worn state and possibly to breakage. Thus the multiclassification of tool states is preferred, which can provide more timely and accurate estimation of tool states. Based on acoustic emission (AE) sensing, this paper proposes a new performance evaluation function for tool condition monitoring (TCM) by considering manufacturing loss. Firstly, two types of manufacturing loss due to misclassification (loss caused by under prediction and loss caused by over prediction) are analyzed, and both are utilized to compute corresponding weights of the proposed performance evaluation function. Then the expected loss of future misclassification is introduced to evaluate the recognition performance of TCM. Finally, a revised support vector machine (SVM) approach coupled with one-versus-one method is implemented to carry out the multiclassification of tool states. With this approach, a tool is replaced or continued not only based on the tool condition alone, but also the risk in cost incurred due to underutilized or overused tool. The experimental results show that the proposed method can reliably perform multiclassificaion of tool flank wear, and reduce the potential manufacturing loss.