Parameter optimization of ultrasonic impact for deformation control based on Dual Information Neural Network

Abstract

Thin-walled structural parts are prone to complex deformations during machining process, primarily due to the residual stress fields within material. Ultrasonic impact treatment is an effective mean for machining deformation control by introducing beneficial compressive stress to counter/cancel the residual tensile stress of the part. However, the effectiveness of residual stress by ultrasonic impact treatment is influenced by various parameters, and a simple treatment of them cannot effectively reduce the deformation of structural parts. Moreover, the initial residual stress field of a part will also influence its overall deformation. In this paper, the influence of different ultrasonic impact size parameters on residual stress was analyzed firstly. A Dual Information Neural Network (DINN) surrogate model was built for rapid deformation prediction by both considering the impact of ultrasonic impact parameters and residual stress within the workpiece, which is further utilized to optimize the ultrasound impact parameters (e.g., impact velocity) through genetic algorithms such that the deformation is minimized. The validation results demonstrate that the proposed DINN can significantly reduce the training time and improve the model prediction accuracy, and the ultrasonic impact parameters selected by the proposed method can effectively reduce the part deformation.