1. Industrial Visual Inspection

Industrial visual inspection technology is considered a "transformative tool" in modern manufacturing and a cutting-edge technology for improving production quality and efficiency. Its purpose is to achieve real-time monitoring and automated detection of product quality in the production process by integrating image acquisition and processing technology, advanced computer vision algorithms, and high-performance artificial intelligence data analysis methods. Industrial visual inspection technology has significantly reduced the reliance on manual inspection, replacing traditional manual inspection methods with accurate and consistent detection capabilities, thereby shortening inspection time, reducing production costs, and improving the reliability and efficiency of the overall production process, promoting the revolutionary transformation of manufacturing towards intelligence and automation.

In response to the key challenges of low abnormal samples, low detection accuracy, poor generalization, and low efficiency in industrial visual inspection tasks, research on industrial defect detection based on 2D/3D fusion machine vision is carried out in key areas such as integrated circuit coating/development, packaging substrate patterning/metallization layers, equipment manufacturing, and steel metallurgy. Independently build the NeuDefect series dataset, including the world's largest packaging substrate micron level defect dataset CPS2D and CPS3D, wafer coating defect dataset WC2D, and cross domain multimodal defect dataset DeMM; Design unsupervised, small sample, zero sample, super-resolution, photometric compensation and other defect detection algorithms, and publish relevant results in top international conferences and academic journals in the fields of computer vision and industrial intelligence such as ICCV, TII, TIM, etc.

Representative achievements:

[1]    R. Yu, B. Guo, and H. Li. "Anomaly Detection of Integrated Circuits Package Substrates Using the Large Vision Model SAIC: Dataset Construction, Methodology, and Application", in IEEE/CVF International Conference on Computer Vision, 2025. (CCF A)

[2]    R. Yu, H. Li, B. Guo and Z. Zhao, "Background-Weaken Generalization Network for Few-Shot Industrial Metal Defect Segmentation", in IEEE Transactions on Instrumentation and Measurement, vol. 74, pp. 1-11, 2025.  (IF:5.9, SCI JCR Q1)

[3]    R. Yu and B. Guo, "Dynamic Reasoning Network for Image-Level Supervised Segmentation on Metal Surface Defect", in IEEE Transactions on Instrumentation and Measurement, vol. 73, pp. 1-10, 2024. (IF:5.9, SCI JCR Q1)

[4]    B. Guo, Y. Wang, S. Zhen, R. Yu, and Z. Su, "SPEED: Semantic Prior and Extremely Efficient Dilated Convolution Network for Real-time Metal Surface Defects Detection", IEEE Transactions on Industrial Informatics, 2023. (IF:11.648, 中科院1区 TOP)

[5]    R. Yu, B. Guo, K. Yang. "Selective Prototype Network for Few-Shot Metal Surface Defect Segmentation", IEEE Transactions on Instrumentation and Measurement, 2022 (71), 1-10. (IF:5.9, SCI JCR Q1)

2. Time Series Perception Data Analysis

Time series perception data analysis technology is one of the core supporting technologies for intelligent manufacturing. It constructs a dynamic monitoring, anomaly warning, and optimization decision-making closed-loop system for manufacturing processes by collecting, processing, and analyzing multidimensional time series data (including sensor signals, equipment status, process parameters, etc.) generated during the production process in real time. This technology deeply integrates time series modeling, machine learning, edge computing and other cutting-edge methods. By extracting key features and deep rules from massive time series data, it significantly improves the intelligence level of the manufacturing system, and effectively solves the long-standing technical bottlenecks in traditional manufacturing industry, such as delayed response, inaccurate fault prediction, and insufficient energy efficiency optimization.

This study focuses on key scientific issues such as multi-source heterogeneous sensor feature fusion, data noise suppression, and mechanism feature analysis faced in temporal perception data analysis tasks. The study focuses on key industrial fields such as high-precision machine tool processing, vehicle manufacturing, and aerospace equipment manufacturing, and systematically conducts a series of research on anomaly detection, fault prediction, quality prediction, and process parameter optimization. During the research process, multiple time-series datasets with industrial application value were constructed, including BMW body welding gun time-series dataset, Foton Cummins machine tool processing dataset, and rotating equipment multi condition fault diagnosis dataset. At the technical level, innovative multi-source heterogeneous data spatiotemporal alignment methods, multimodal feature interaction mechanisms, and multi-scale analysis frameworks have been proposed for industrial scenarios. Relevant research results have been published in authoritative journals in the field of industrial intelligence such as Computers in Industry and Neurocomputing, providing important theoretical support and technical solutions for intelligent manufacturing.

Representative achievements:

[1]    M. Chen, R. Yu, Z. Liang, K. Li, and H. Qi. "A multiscale process-aware retention network for fault prediction in mixed-model production", in Computer in Industry, 2025. (IF:8.9, 中科院1区 TOP)

[2]    R. Yu, M. Chen, and B. Liu, "A triple-phase boost transformer for industrial equipment fault prediction", in Neurocomputing, 2024.  (IF:6.5, SCI JCR Q1)

[3]    R. Yu and B. Guo, "Is multi-level data enhancement helpful for knowledge graph? A new perspective on multimodal fusion", in Knowledge-Based Systems, 2024. (IF:7.6, 中科院1区 TOP)

[4]    K. Yang, R. Yu, B. Guo, and J. Li, "Interaction Subgraph Sequential Topology-Aware Network for Transferable Recommendation",  in IEEE Transactions on Knowledge and Data Engineering, 2024. (CCF A)

3. 3D Spatial Intelligence

3D spatial intelligence refers to the ability of a system to understand, perceive, and reconstruct the environment in three-dimensional space. This technology has helped autonomous driving and robots gain the ability to observe and perceive the world. By integrating multimodal sensor data such as LiDAR, cameras, and millimeter wave radar, and using advanced artificial intelligence algorithms to analyze and reconstruct the data, 3D spatial intelligence enables autonomous vehicles and robots to achieve efficient and reliable environmental perception and reconstruction in dynamic and changing environments, thereby ensuring their autonomy and safety.

To address the challenges of complex working environments and low safety redundancy in autonomous driving and robotics, we aim to use 3D spatial intelligence to assist mobile units in accurately perceiving surrounding obstacles, lane markings, traffic signs, and other vehicles and pedestrians. We will also construct high-precision 3D maps in real-time and complete safe driving path planning. For robots, whether it is industrial robots performing precise operations or service robots navigating in complex environments, 3D spatial intelligence enables them to accurately recognize objects, understand scene layouts, and autonomously complete tasks. The relevant achievements have been published in journals such as ICAI, DCAN, TIM, etc.

Representative achievements:

[1]   S. Cao, J. Li, and R. Yu, "DWT-3DRec: DeepJSCC-based wireless transmission for efficient 3D scene reconstruction using CityNeRF", in Digital Communications and Networks, 2025.(IF:7.5, JCR Q1) 

[2]   Z. Cao, T. Wang, P. Sun, F. Cao, S. Shao, and S. Wang, "ScorePillar: A Real-Time Small Object Detection Method Based on Pillar Scoring of Lidar Measurement", in IEEE Transactions on Instrumentation and Measurement, vol. 73, pp. 1-13, 2024.(IF:5.9, SCI JCR Q1)

[3]  J. Li, Y. Zhang, L. Wang, and R. Yu, "SA4D-HDR: Segment Anything with Neural Radiance Fields for 4D HDR Scenes", in European Conference on Artificial Intelligence 2025.(CCF B)