TOP

The research group will recruit several PHD and master students and 2 postdocs.

There are three requirements for doctoral and master students  enrollment: (1) Applicants should have an engineering background and have a strong interest in information technology. Applicants should have obtained a relevant bachelor's or master's degree; (2) Strong technical background, including but not limited to research experiences in offshore engineering structural analysis, marine environment numerical simulation and prediction, marine digital twin, etc. Candidates with research or practical experiences in numerical algorithms, industrial software development, or high-performance computing will be preferred; (3) Highly self-motivated, good written and oral English communication skills, and independent working ability.

Postdoctoral recruitments should also meet the following two points: (1) The applicant should be under the age of 35 and have obtained a doctoral degree no more 3 years; (2) The research directions are civil engineering information technology, Marine environmental information modeling and application, data-driven knowledge discovery and application, etc. (Note: postdoctoral candidates are required to present a half-hour academic presentation, including the main research works during PHD period and future postdoctoral work plans).

If you are interested, please send your resume, transcripts and work plan to the email: hu.zhenzhong@sz.tsinghua.edu.cn. For details, please see: PHD Master Recruitment and Postdoctoral Recruitment.

2026-09-18

The paper "Fine-tuning vision foundation model for crack segmentation in civil infrastructures" has received over 100 citations on Google Scholar. It was published in May 2024 and achieved over 100 citations within just over two year, which is the fastest rate of citation breakthrough to date.

The rapid growth in citations closely reflects the broader trend of intelligent transformation in civil engineering. Traditional vision-based detection models trained on small datasets commonly suffer from limited generalization and poor adaptability to complex working conditions, which has constrained the large-scale application of intelligent inspection technologies in engineering practice. This study introduced a vision foundation model for crack segmentation and adapted it through parameter-efficient fine-tuning, effectively addressing these bottlenecks and providing a technical paradigm that can be directly adopted by related research.

The techniques adopted in the paper can also be extended to segment other types of structural defects, such as earthquake damage segmentation, rebar corrosion segmentation, and structural water seepage identification, gradually covering various inspection needs in structural health monitoring (SHM). Building on this technical framework, the team will carry out further exploration to advance the deployment of AI technologies in front-line engineering inspection and maintenance, providing strong support for the intelligent management of civil infrastructure.

2026-09-11

The paper "Beyond annual dose: Assessing lifetime cancer risk from dietary radiocesium exposure and public risk-perception gaps in Japan" has been published in Environment International.

Radiocesium in food poses an ongoing challenge in environmental health risk assessment. Annual dose compliance (1 mSv/year) is essential for radiological protection but may not reflect cumulative lifetime cancer risk (LCR) under chronic exposure or corresponding public risk perceptions. This study developed an integrated radiological risk assessment framework. Using a nationwide survey in Japan together with food-monitoring data on radiocesium concentrations, prefecture-level LCR and risk-perception gaps were estimated through multilevel regression with poststratification. The annual dietary radiation dose was well below the 1 mSv/year benchmark, while modeled mortality and morbidity LCRs remained above 10⁻⁵, ranging from 1.84×10⁻⁵ to 1.37×10⁻⁴ and from 2.70×10⁻⁵ to 2.02×10⁻⁴, respectively. The high-risk prefectures identified in spatial analyses, particularly in Tohoku, warrant prioritized monitoring. Agricultural products were the dominant exposure pathway, accounting for approximately 55.0–64.5% of the estimated LCR. Estimated risk tended to exceed perceived risk in northern Japan, indicating a need for proactive risk control, whereas perceived risk tended to exceed estimated risk in central Japan, suggesting a greater need for targeted communication. By considering the LCR alongside annual dose metrics, this framework supports region-specific monitoring and risk communication.

Note: Environment International is a Q1-ranked TOP journal in environmental sciences and ecology, with a 2026 impact factor of 10.2. Professor Mao Liang and Professor Huang Lei from Nanjing University are the corresponding authors. This research was supported by the National Natural Science Foundation of China.

2026-09-01

On September 1, Dr. Lin Jiarui, Associate Researcher and Director of the Digital Construction Teaching Laboratory at Tsinghua University, was invited to our institute to deliver an academic lecture titled "Key Technologies for Intelligent Engineering Design Based on BIM and Development of Independent Software."

Engineering design is a critical stage that significantly influences construction quality and performance. Enhancing design quality has long been a focal point in the engineering field. In his talk, Dr. Lin shared his team's latest research advances in intelligent BIM-based design, including modular BIM generation, AI-enhanced design simulation, the QwenBIM large model, intelligent BIM review, and defect repair technologies, along with their practical applications. Additionally, he discussed progress in developing independent BIM software through case studies on three-dimensional structural design for offshore fixed platforms, exploring integrated solutions for intelligent engineering design. The presentation, both cutting-edge and practical, featured rich real-world examples and sparked enthusiastic discussions and in-depth exchanges among faculty and students.

Note: Dr. Lin Jiarui has long been engaged in research on intelligent drawing review, intelligent inspection, digital twins, and large-scale engineering models. He has led and participated in multiple national-level projects, including those funded by the National Natural Science Foundation of China and the Key R&D Program. He has published over 150 academic papers, co-authored nine industry and local standards, and holds more than 30 authorized patents and software copyrights. He has been selected for the China Association for Science and Technology’s “Young Talent Support Program,” and has received numerous awards, including a Gold Medal at the Geneva International Invention Exhibition, the Champion Award at the bSI openBIM Competition, the First Prize of Huaxia Construction Science and Technology, the Special Award for Scientific and Technological Progress from the Yellow River Conservancy Commission, the First Prize for Teaching Achievements at Tsinghua University, and Tsinghua University’s “Outstanding Teacher and Friend” Award.

2026-08-30

The paper, titled "Recovering layer information for precise component segmentation from 2D drawings", has been published in Advanced Engineering Informatics.

2D drawings offer a cost-effective data source for BIM reconstruction of existing buildings, yet automated component segmentation remains challenging due to missing layer information in raster drawings. To address the problems of imprecise boundaries in existing methods and the semantic ambiguity caused by neglecting line-symbol relationships, this study proposes a two-module framework: first, a Graph Convolutional Network (GCN)-based method that incorporates line-symbol proximities into a heterogeneous graph for layer recovery; and second, a rule-based method that reconstructs vectorized component contours from the recovered layers, while resolving incomplete contours caused by dashed lines and layer recovery errors. On residential building drawings, the framework achieves an F1 score of 84.3% for layer recovery and a mean IoU of 80.1% for component segmentation, with improved boundary consistency over CNN-based baselines. Further experiments on office building drawings and paper drawings demonstrate the framework's applicability. A preliminary end-to-end BIM modeling test shows that the generated contours require less boundary post-processing than CNN-based masks, providing a practical foundation for BIM reconstruction of existing buildings.

Note: Advanced Engineering Informatics is a Q1-ranked top journal in the field of engineering and technology. Its 2026 impact factor was 11.5. The first author of the paper is Lu Sun, a doctoral student from the Hong Kong University of Science and Technology. Yu Yantao from the Hong Kong University of Science and Technology is the corresponding author. The research results were funded by the Innovation and Technology Fund Seed Research Project, the RGC Theme-based Research Scheme, the Guangdong NSF, and the Guangdong Basic and Applied Basic Research Foundation.