A mechanism informed neural network for predicting machining deformation of annular parts. (August 2022)

• Record Type: Journal Article
• Authors: Ni, Yang; Li, Yingguang; Liu, Changqing; Liu, Xu
• Is Part Of: Advanced engineering informatics. Volume 53(2022)
• Publication Date: 2022-08
• Subjects: Annular part -- Machining deformation prediction -- Data-mechanism fusion -- Thin-shell theory -- Fourier series; Computer-aided engineering -- Periodicals; Engineering -- Data processing -- Periodicals; 620.00285
• DOI: 10.1016/j.aei.2022.101661
• Languages: English
• ISSNs: 1474-0346

Abstract

Graphical abstract: Highlights: A mechanism informed neural network (MINN) is proposed for deformation prediction. Deformation mechanism model is established by Thin-shell theory and Fourier series. MINN is built with a new structure and loss function guided by the mechanism model. MINN can be used to predict the annular part deformation accurately and stably. Better results are obtained by MINN compared with other data-driven methods. Abstract: Controlling machining deformation of annular parts is crucial for ensuring the performance of high value products and equipment. For example, during manufacturing of critical parts in aircrafts and spacecrafts, accurate prediction of machining deformation is the basis for guiding the formulation of deformation control strategies. However, due to the complexity of the machining deformation of annular parts, existing methods still have limitations in accurate prediction. To this end, this paper proposes a mechanism informed neural network (MINN) to predict machining deformation of annular parts. MINN is realized by establishing the dual sub-networks structure and using enhanced loss functions with the consideration of the deformation mechanism model characteristics of annular parts. The deformation was decomposed into the axisymmetric portion and the non-axisymmetric portion according to the deformation superposition principle, and modeled separately based on the thin-shell theory and Fourier series. Experiment results showed that the proposed method could predict the machining deformation of annular parts more accurately and stably with a small amount of training data, compared with previous methods.