CNC internal data based incremental cost-sensitive support vector machine method for tool breakage monitoring in end milling

Abstract

A tool breakage monitoring (TBM) system needs to detect tool breakage promptly in an unattended automation workshop. Traditional TBM systems that employ external sensors to acquire diagnostic signals such as spindle power for making judgments are inconvenient since extra sensors should be installed. Moreover, the signals from the external sensors are independent of the computer numerical control (CNC) system, and it is difficult to label them with the corresponding machining task information so that the target data can be segmented automatically. This paper proposes an incremental cost-sensitive support vector machine (ICSSVM) tool breakage monitoring method based on CNC internal data, which is inherently machining-task labeled and can be accessed directly from the CNC system without extra sensors. To satisfy the dataset’s integrity at the initial stage of model training, a simulation method that is based on the actual tool breakage characteristics is applied to generate simulated tool breakage data. The ICSSVM method combines cost-sensitive SVM (CSSVM) and modified incremental SVM (ISVM) to solve the imbalanced classification problem, which increases the misclassification probability of the minority class, train the model incrementally from the absence of samples, and guarantee high algorithm efficiency as the size of the dataset increases. It is proved that the ICSSVM algorithm has better algorithmic efficiency compared to the batch cost-sensitive SVM (BCSSVM). It is also proved that the ICSSVM algorithm has better imbalanced classification performance than batch SVM (BSVM), as assessed by the receiver operator characteristic (ROC) curves. The industrial practicability of the proposed method is verified by actual machining with a CNC system integrated with the TBM module.