文献类型：期刊文献

英文题名：Predicting submerged deflecting abrasive waterjet peening induced surface roughness based on vibration signal via a time-frequency parallel deep learning network

作者：Chi, Yu-Xin[1];Yao, Shu-Lei[1];Zhu, Xian-Hao[1];Wang, Zhi-Yun[1];Qi, Jie[1];Wang, Ning[1];Zhang, Xian-Cheng[1]

机构：[1]East China Univ Sci & Technol, Key Lab Pressure Syst & Safety, Minist Educ, Shanghai 200237, Peoples R China

年份：2026

卷号：161

起止页码：424

外文期刊名：JOURNAL OF MANUFACTURING PROCESSES

收录：;EI(收录号：20260620029102);WOS:【SCI-EXPANDED(收录号:WOS:001689239400001)】；

基金：This work was financially supported by the National Key Research and Development Program (No. 2022YFB4600019) , the Postdoctoral Fellowship Program of CPSF (No. GZB20240219) , and the Shanghai Sailing Program (No. 24YF2708100) .

语种：英文

外文关键词：Submerged deflecting abrasive waterjet; peening; Vibration signal; Surface roughness; Prediction; Deep learning

摘要：Waterjet peening (WJP) is an effective surface strengthening method for complex aeroengine components. However, the low stiffness of thin-walled parts, such as blades, can induce significant vibration responses during machining, which may adversely affect the final surface integrity. Therefore, process monitoring is crucial for capturing dynamic responses and providing timely feedback for quality control. However, owing to the complex working environment and strengthening mechanism, monitoring the WJP process remains challenging. In this study, vibration signals from thin-walled titanium alloy TA19 specimens were collected during submerged deflecting abrasive waterjet peening (SDAWJP). The vibration signals exhibited change trends consistent with those of surface roughness and showed clear sensitivity to variations in process parameters, thus enabling accurate prediction modeling. A time-frequency parallel deep learning network (TFPNet) was proposed, in which vibration features from both the time and frequency domains were extracted simultaneously and subsequently fused via an attention pooling mechanism to effectively predict the surface roughness. The prediction performance was evaluated under varying abrasive flow rates and water pressures. The proposed model achieved mean absolute percentage errors of 0.99% and 1.35%, respectively, which were significantly lower and more stable than those of the comparative methods. The findings are expected to provide support for the machining quality assurance in practical aviation components.