Deep learning-based instantaneous cutting force modeling of three-axis CNC milling. (15th May 2023)

• Record Type: Journal Article
• Authors: Xie, Jiejun; Hu, Pengcheng; Chen, Jihong; Han, Wenshuai; Wang, Ronghua
• Is Part Of: International journal of mechanical sciences. Volume 246(2023)
• Publication Date: 2023-05-15
• Subjects: CNC milling -- Cutting force -- Deep learning -- Data-driven modeling -- Interpretation of deep learning model; Mechanical engineering -- Periodicals; Génie mécanique -- Périodiques; Mechanical engineering; Maschinenbau; Mechanik; Zeitschrift; Periodicals; 621.05
• DOI: 10.1016/j.ijmecsci.2023.108153
• Languages: English
• ISSNs: 0020-7403

Abstract

Highlights: Deep learning-based instantaneous cutting force model of CNC milling is presented. Comprehensive geometric and processing information is encoded into images. Deep learning network named MF-CNN is designed to predict the cutting force. Interpretation of MF-CNN is analyzed toward the theoretical cutting force model. Proposed approach can predict cutting force with outstanding accuracy. Abstract: Accurate cutting force modeling is the basis for good planning and optimization of the process and parameter of Computerized Numerical Control (CNC) milling. Traditional cutting force prediction models suffer from problems of oversimplifications on the model's input and framework, making it difficult to predict the cutting force accurately in the complex machining process. This paper proposes a novel deep learning-based instantaneous cutting force prediction model with superior modeling precision. According to the mechanism of cutting force generation, the comprehensive geometric and processing information in the machining process is creatively expressed as multi-channel digital images named Image of Comprehensive Geometric Processing Information (ICGPI). A deep learning network called Milling Force Convolutional Neural Network (MF-CNN) is then designed that takes the ICGPI as the input and the three-dimensional instantaneous cutting forces as the output. To address the challenging problem of interpretation of the deep learning network, the MF-CNN is analyzed toward the theoretical mechanistic cutting force model, validating that the proposed method can fully cover all the geometric information and mathematical operations involved in the theoretical model. Finally, some physical cutting experiments are conducted to validate the effectiveness and superiority of the proposed method, showing that our MF-CNN can predict the instantaneous cutting force with outstanding accuracy and is much superior to the three most popular benchmarks. Graphical abstract: Image, graphical abstract