Application of Deep Learning to Predict Shot Peening Coverage

Author:  Y.H.A Chua1, Z.B. Wang1, H.C. Ang1, and A. Shukri1 1. Advanced Remanufacturing and Technology Centre, 3 Cleantech Loop, #01/01 CleanTech Two, Singapore 637143
Source:  ICSP14 Milan
Doc ID:  2022066
Year of Publication:  2022
Abstract:  
Coverage is one of the key process control variables for the shot peening process. Currently, the method to quantify coverage relies heavily on human visual judgement, which is highly subjective. This study explored the use of deep learning to predict coverage using images captured from peened SS316 and Ti-6Al-4V materials. It was found that all deep learning models developed with the SuaKIT software could achieve a high prediction accuracy of at least 93% based on fixed intensity and material testing. However, the models performed poorly when tested using crossed datasets with different materials and intensities. Combining the data from different datasets can improve the prediction accuracy to at least 93.5%. Keywords: Shot Peening, Coverage, Deep Learning