Enhancing fidelity of mechanistic cutting force model using hybrid data mining approach

Cutting force is a vital indicator for assessing the performance of machining operations; therefore, developing a reliable predictive model is mandatory for process monitoring, optimization, and control. The mechanistic force model is preferred among approaches presented in the literature due to computational efficiency and effectiveness. The prediction abilities of the model largely depend on the empirical relationship between uncut chip geometry and lumped empirical constants determined using machining experiments involving cutting force measurements. The experimental data usually contains noise and outliers that must be removed before determining empirical constants. Data mining techniques are potential tools for removing noisy and outlier-ridden experimental data. This work develops a two-stage hybrid approach combining a clustering technique and machine learning model for data mining and determining precise empirical constants. The comparative evaluation of three unsupervised clustering techniques, namely hierarchical density-based spatial clustering of applications with noise (HDBSCAN), one-class support vector machine (SVM), and elliptic envelope, is performed in removing outliers. The machine learning-based ADAMW algorithm is implemented to fit the relationship and determine empirical constants from cleansed experimental data. It has been shown that HDBSCAN combined with the ADAMW algorithm can effectively remove outliers, enhance the goodness of fit, and achieve better performance. The prediction abilities of the proposed approach are corroborated by performing machining experiments over varying cutting conditions. It is concluded that the hybrid approach can address challenges associated with noisy and outlier-ridden experimental data, thereby enhancing the fidelity of the mechanistic force model.