Tool-tip vibration prediction based on a novel convolutional enhanced transformer

Abstract

Superior surface finish remains a fundamental criterion in precision machining operations, and tool-tip vibration is an important factor that significantly influences the quality of the machined surface. Physics-based models heavily rely on assumptions for model simplification when applied to complex high-end systems. However, these assumptions may come at the cost of compromising the model's accuracy. In contrast, data-driven techniques have emerged as an attractive alternative for tasks such as prediction and complex system analysis. To exploit the advantages of data-driven models, this study introduces a novel convolutional enhanced transformer model for tool-tip vibration prediction, referred to as CeT-TV. The effectiveness of this model is demonstrated through its successful application in ultra-precision fly-cutting (UPFC) operations. Two distinct variants of the model, namely, guided and nonguided CeT-TV, were developed and rigorously tested on a data set custom-tailored for UPFC applications. The results reveal that the guided CeT-TV model exhibits outstanding performance, characterized by the lowest mean absolute error and root mean square error values. Additionally, the model demonstrates excellent agreement between the predicted values and the actual measurements, thus underlining its efficiency and potential for predicting the tool-tip vibration in the context of UPFC.