Keynote Speakers

Prof. Zhenghao Shi

Xi'an University of Technology, China

IEEE Senior Member

Biography: Zhenghao Shi, Ph.D., Professor, Doctoral Supervisor, Member of the Academic Committee of Xi'an University of Technology, Outstanding Member of CCF, IEEE Senior Member, "500 Elite Talents" of Taizhou City, Zhejiang Province, Executive Director of Shaanxi Computer Society, Chairman of the "Computer Vision Technology Professional Committee" of Shaanxi Computer Society, Vice Chairman of the "Biomedical Intelligent Computing Professional Committee" of Shaanxi Computer Society, Head of the "Intelligent Image Processing and Application" research team at Xi'an Technological University, with main research directions in machine vision, medical image processing, and machine learning. Published 60 academic papers as first author or corresponding author, authorized 15 invention patents (including 1 South African invention patent), won 2 second prizes of Shaanxi Science and Technology Progress Award (ranked first), 1 second prize of Xi'an Science and Technology Progress Award (ranked first), and 3 second prizes of Shaanxi Higher Education Science and Technology Award (ranked first). Firstly, One second prize for scientific and technological progress from Shaanxi Computer Society (ranked first), one first prize for technological invention from Shaanxi Computer Society (ranked first), and the "Wiley China Open Science High Contribution Author" award.

Prof. Shiyuan Wang

Southwest University, China

IEEE Senior Member

Biography: Shiyuan Wang, IEEE Senior Member, Southwest University “Signal and Intelligent Information Processing” research team leader, Senior Member of the Chinese Institute of Electronics, Committee Member of the University Electronic Technology Specialized Committee of the Chongqing Institute of Electronics. He was a Research Associate at The Hong Kong Polytechnic University in 2013 and a Research Fellow at The City University of Hong Kong in 2023, respectively. His research focuses on adaptive signal processing, integrated navigation, nonlinear dynamics, and simultaneous localization and mapping (SLAM). He has published one monograph and over 150 papers in leading journals and conference proceedings, including one ESI Highly Cited Paper. He has been awarded the Third Prize of the Chongqing Natural Science Award (ranked third) and has led over 10 research projects, including four funded by the National Natural Science Foundation of China. From 2018 to 2021, he served as an Associate Editor for IEEE TCAS II, a leading journal in circuits and systems, and currently serves as an Associate Editor for Symmetry.

Ttile: Neural Network-Aided Kalman Filtering: Combining the Strengths of Model-Based and Data-Driven Methods

Abstract: Accurate real-time estimation of latent states is a fundamental task in signal processing, with applications in intelligent industrial systems, precise localization, and reliable navigation. Among existing approaches, the Kalman filter (KF) and its variants—representative model-based (MB) methods that rely on well-defined state-space models (SSMs)—have been widely adopted for their low complexity and fast convergence. In this talk, we will first reviewing the MB methods, including their applications and limitations. However, in complex and uncertain scenarios where accurate models are hard to build, the performance of these MB methods drops considerably. To this end, the model-agnostic nature of neural networks has spurred the development of data-driven (DD) methods for state estimation in uncertain dynamics by learning from data. These emerging DD methods still require large amounts of labeled data for effective training and lack robustness to SSM fluctuations. To address these challenges, hybrid methods have been proposed with the aim of leveraging the strengths of both MB and DD approaches while compensating for their respective weaknesses. By replacing the computation of intermediate values in MB framework which are affected by inaccurate SSM information with neural networks, these hybrid methods are capable of utilizing the strong nonlinear modeling capability of neural networks to handle highly nonlinear and inaccurate SSMs, while alleviating the need for extensive labeled data since networks are only trained to approximate intermediate values. Therefore, supported by extensive evidence, these hybrid methods position themselves as a premier strategy for future research in robust and data-efficient state estimation, addressing the long-standing challenges of complexity and uncertainty.

Prof. Yajun Liu

South China University of Technology, China

Biography: Yajun Liu, Ph.D., Professor, is affiliated with the School of Mechanical and Automotive Engineering at South China University of Technology. He is a member of the American Society of Mechanical Engineers (ASME), the Society of Automotive Engineers (SAE), a senior member of the Chinese Mechanical Engineering Society, and a committee member of the Manufacturing Technology Professional Committee of the Chinese Automation Society. He also serves as a technical consultant for multiple leading enterprises in the industry. His research focuses on the mechanisms of manufacturing system processes and optimization control technologies. He has received numerous academic awards, including the “Nanyue Outstanding Graduate and Zeng Xianzi Scholarship First Prize,” the “Elianda Innovation Award,” the “Modern Manufacturing Academic Paper Award,” and the “Excellent Paper Award at the 20th International Conference on Mechatronics in 2016,” among others. Professor Liu has published over 150 papers in professional journals and international conferences in the fields of machining mechanics, mechanical system process mechanisms, and optimization control research, as well as more than 10 papers on scientific teaching research. He has led and undertaken more than 10 national and provincial-level projects, in addition to over 20 enterprise R&D projects.

Ttile: Neural Network Agent Model for Hydraulic Impact Hammer

Abstract: Through the high-fidelity digital simulation model of hydraulic hammer, the energy conversion rate of hydraulic system of the pile hammer can be calculated. With the help of the calculation results, the post-processing program can match the control parameters of the pile machine with the working conditions to maximize the energy conversion efficiency. However, the application of this method is limited by the tight computing power in pile hammer equipment. Therefore, it is necessary to generate a surrogate model, which should meet the following conditions: perform accurate calculations and replace simulation software, able to calculate the energy conversion rate of the hydraulic system in the construction process of the hydraulic pile hammer, and operate efficiently in the equipment with limited computing power. Using neural network to fit the calculation results of simulation software can greatly reduce the calculation cost of prediction process. However, the unexplained feature of neural network output increases the application risk of this method. Based on the classical theory of physics-informed neural network (PINN), a PINN method based on inequalities is proposed in this paper. Based on this method, a physics-informed surrogate model network (PISMN) oriented to the simulation process of hydraulic pile hammer was constructed. It is proved that this method can constrain the output of the surrogate model, improve the stability of the training process, and the median prediction deviation of the prediction results was reduced by 53.0% in the validation set with perturbation.