Tool wear is a detrimental circumstance in end milling and estimating its occurrence in machinery is an onerous process. Indirect tool monitoring has been actively studied to identify instances of wear on the cutting tool based on the signal from a sensor that represents the tool condition. Runout of a machine spindle during machining as a result of a defective tool commonly occurs in the metal cutting process. In this study, gap sensors were installed at the machine spindle to measure the runout. Two types of tool conditions and four cutting depths were considered during end milling to identify the relation between the spindle runout, cutting depth, and tool condition based on the gap sensor signal. Statistical features were extracted from the signals obtained, and a feature selection technique was applied to identify the ideal features as an input for the machine learning (ML) algorithms, specifically support vector machine (SVM) and multi-layer perceptron neural network (MLP NN). The SVM models were evaluated through k-fold cross-validation, while stochastic learning was applied to the MLP NN models to obtain the most compatible algorithm for the binary classification. The performance of SVM and MLP NN algorithms in classifying the signal based on the tool condition was studied and compared. The SVM outperformed the MLP NN in terms of classification accuracy, F1-score, precision, and sensitivity for all datasets despite the minimal parameter assignment in the former.

刀具磨损在立铣刀中是有害的情况，估计在机械中的磨损是一个繁重的过程。已经积极研究了间接工具监视，以根据代表工具状态的传感器信号识别切削工具的磨损情况。在加工过程中，由于刀具故障导致机床主轴跳动通常发生在金属切削过程中。在这项研究中，间隙传感器安装在机床主轴上以测量跳动。在端铣削过程中考虑了两种类型的刀具状态和四种切削深度，以根据间隙传感器信号确定主轴跳动，切削深度和刀具状态之间的关系。从获得的信号中提取统计特征，然后应用特征选择技术来识别理想特征，作为机器学习（ML）算法的输入，特别是支持向量机（SVM）和多层感知器神经网络（MLP NN）。通过k倍交叉验证对SVM模型进行评估，同时将随机学习应用于MLP NN模型以获得最兼容的二进制分类算法。研究并比较了SVM和MLP NN算法在基于工具条件的信号分类中的性能。尽管在前者中分配的参数最少，但SVM在分类精度，F1得分，精度和灵敏度方面都优于MLP NN。通过k倍交叉验证对SVM模型进行评估，同时将随机学习应用于MLP NN模型以获得最兼容的二进制分类算法。研究并比较了SVM和MLP NN算法在基于工具条件的信号分类中的性能。尽管在前者中分配的参数最少，但SVM在分类精度，F1得分，精度和灵敏度方面都优于MLP NN。通过k倍交叉验证对SVM模型进行评估，同时将随机学习应用于MLP NN模型以获得最兼容的二进制分类算法。研究并比较了SVM和MLP NN算法在基于工具条件的信号分类中的性能。尽管在前者中分配的参数最少，但SVM在分类精度，F1得分，精度和灵敏度方面都优于MLP NN。