Yaoyu Ding

Dr. Yaoyu Ding


Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
21 December 2024 in Optics & Laser Technology
Verified
Journal Article
Closed access
JLJinjian Li
Jinjian Li
JZJinhan Zhao
Jinhan Zhao
YLYuxin Luo
Yuxin Luo
JHJiaqian Huang
Jiaqian Huang
YDYaoyu Ding
Yaoyu Ding

20 September 2024 in Sensors
Verified
Journal Article
Open access
PDF file
SGShouwei Gao
Shouwei Gao
YZYali Zheng
Yali Zheng
SLShengping Li
Shengping Li
JZJie Zhang
Jie Zhang
LBLi Bai
Li Bai
YDYaoyu Ding
Yaoyu Ding
In this paper, we focus on eddy current array (ECA) technology for defect detection in finely grooved structures of spinning cylinders, which are significantly affected by surface texture interference, lift-off distance, and mechanical dither. Unlike a single eddy current coil, an ECA, which arranges multiple eddy current coils in a specific configuration, offers not only higher accuracy and efficiency for defect detection but also the inherent properties of space and time for signal acquisition. To efficiently detect defects in finely grooved structures, we introduce a spatiotemporal self-attention mechanism to ECA testing, enabling the detection of defects of various sizes. We propose a Multi-scale SpatioTemporal Self-Attention Network for defect detection, called MSTSA-Net. In our framework, Temporal Attention (TA) and Spatial Attention (SA) blocks are incorporated to capture the spatiotemporal features of defects. Depth-wise and point-wise convolutions are utilized to compute channel weights and spatial weights for self-attention, respectively. Multi-scale features of space and time are extracted separately in a pyramid manner and then fused to regress the bounding boxes and confidence levels of defects. Experimental results show that the proposed method significantly outperforms not only traditional image processing methods but also state-of-the-art models, such as YOLOv3-SPP and Faster R-CNN, with fewer parameters and lower FLOPs in terms of Recall and F1 score.

8 August 2024 in Photonics
Verified
Journal Article
Open access
PDF file
JZJinhan Zhao
Jinhan Zhao
CCChao Chao
Chao Chao
YDYaoyu Ding
Yaoyu Ding
LBLi Bai
Li Bai
YCYuhua Cheng
Yuhua Cheng
Laser drilling is widely used for fabricating holes in the semiconductor industry due to high throughput and a small heat-affected zone. However, it produces varying depths owing to uncertain external conditions and requires live control at the rate of a few tens of kHz to handle the fast material removal rate. Optical coherent tomography is capable of in situ acquiring a raw interferogram at a high rate (>80 kHz), but the depth extraction is slow due to the involved heavy Fast Fourier Transform (FFT). To address this, an efficient depth-tracking algorithm is proposed to save the FFT. It searches the depth in the raw interferogram locally with a known last depth given the two truths that only one depth exists and the adjacent depths do not change significantly. The proposed algorithm was proven to expedite the measuring rate six times with sub-pixel tracking precision. To further secure the rate against the interrupting of the system, the tracking process is parallelly implemented in a field-programmable gate array. The closed-loop control tests were conducted on probe cards with depth variations introduced by offsetting laser focus. The proposed method maintained a uniform depth, with variations reduced by 80% compared to traditional methods.

11 December 2025 in Materials
Verified
Journal Article
Open access
PDF file
CXChunyang Xia
Chunyang Xia
KZKui Zeng
Kui Zeng
JNJiawei Ning
Jiawei Ning
YDYaoyu Ding
Yaoyu Ding
YLYonghui Liu
Yonghui Liu
Laser–arc hybrid additive manufacturing (LAHAM) combines the benefits of arc-based deposition and laser precision but involves complex, nonlinear process interactions that challenge the prediction and control of bead geometry and energy consumption. This study develops a machine learning (ML) framework to predict bead width, height, and Deposition volume per unit energy (DVUE) in LAHAM. Using experimental data, multiple regression models—including Support Vector Regression, Gaussian Process Regression, Neural Networks, and XGBoost—were trained and evaluated. Gaussian Process Regression (GPR) demonstrated superior performance in capturing nonlinear relationships and was further optimized using Bayesian Optimization and Particle Swarm Optimization. The optimized GPR models were integrated with the NSGA-II multi-objective optimization algorithm to simultaneously minimize geometric deviations and maximize DVUE. Results show that the proposed approach effectively identifies Pareto-optimal process parameters, achieving a balance between deposition accuracy and energy utilization rate, thereby providing a reliable and intelligent strategy for process optimization in hybrid additive manufacturing.

27 October 2026 in Photonics
Verified
Journal Article
Open access
PDF file
YWYingjian Wang
Yingjian Wang
BWBixia Wang
Bixia Wang
LXLang Xu
Lang Xu
XLXiaolin LIU
Xiaolin LIU
QCQi Chen
Qi Chen
YDYaoyu Ding
Yaoyu Ding
Laser welding of hairpin ends is a key joining process for electric-vehicle (EV) traction motors, but pre-weld geometric deviations—including lateral gap variation, axial height mismatch, and focus offset—can compromise weld stability and long-term motor reliability. Conventional coaxial camera inspection provides reliable two-dimensional information but cannot directly measure axial mismatch or local end-face topography. To address this limitation, this study applies a coaxial in-line inspection system that integrates a spectral-domain optical coherence tomography (OCT) depth-sensing path and a CMOS visual imaging channel into a common optical axis. Unlike conventional OCT applications that aim for multilayer subsurface imaging in transparent media, the OCT channel in this work profiles the single reflective interface of opaque copper end faces and achieves axial linear residual errors within ±15 μm over a 4.5 mm working range, given an approximately 900 mm focus distance. The visual channel locates the weld seam and evaluates in-plane longitudinal stagger, whereas the OCT channel performs fixed-position B-scan profiling for lateral gap, axial height mismatch, end-face topography defects, and focus-offset assessment without requiring vision-guided scan repositioning. Experiments on 16 hairpin-end mock-up pairs, including 13 intentionally introduced defect samples, achieved correct classification for all tested samples. The OCT depth-profiling step was completed within the visual inspection time window, enabling parallel OCT–visual inspection without extending the original production takt time. These results demonstrate the feasibility of this non-contact method for pre-weld quality assurance in the EV stator laser welding scenario.

27 June 2025 in IEEE Transactions on Industrial Informatics
Verified
Journal Article
Closed access
YZYali Zheng
Yali Zheng
ZCZhiming Chen
Zhiming Chen
LWLixun Wu
Lixun Wu
LBLi Bai
Li Bai
YDYaoyu Ding
Yaoyu Ding
Graphs are widely recognized for their robust ability to represent objects through nodes and edges. This article aims to provide stable visual guidance for operators to identify electronic components on printed circuit board (PCB) images during the manufacturing and maintenance processes, using graph techniques to enhance the efficiency of component localization. A three-stage approach is proposed to achieve this objective. First, a method based on graph neural networks is introduced for electronic component detection, trained exclusively on the public FISC-PCB dataset. Second, the registration of the physical location with PCB images is modeled as a Markov Chain, and a fast subgraph matching algorithm is proposed to reduce time and space complexities by leveraging a second-order attribute similarity approximation. Finally, the random sample consensus algorithm is employed to eliminate outliers and calculate geometric relationships, enabling accurate positioning. The proposed framework is evaluated using ten PCB images captured by line-scan cameras or mobile phone cameras and compared with leading contemporary methods. The various experimental results demonstrate that the proposed method outperforms existing state-of-the-art solutions in localization accuracy and meets the requirements of efficiency and space utilization.