The International Centre for Numerical Methods in Engineering (CIMNE) has launched DAMSHAI (Dam Structural Health Monitoring and Safety Assessment with an AI Agent), a three-year research project that will explore the application of artificial intelligence to critical infrastructure safety assessment, addressing growing demands for more efficient and objective prevention.
With a budget of €126,250 and funded by Spain’s Ministry of Science, Innovation and Universities, the project is led by Professor Fernando Salazar from the centre’s Machine Learning and Models in Hydro-Environmental Engineering, and will run until August 2028.
Testing AI’s Role in Dam Safety
DAMSHAI will evaluate whether artificial intelligence agents can effectively interpret complex monitoring data sets and provide decision support to engineers responsible for dam safety. While numerical models and machine learning techniques have advanced the prediction of dam behaviour, safety assessments still depend heavily on time-intensive expert analysis.
The project will test three distinct AI-based approaches: existing Large Language Models (LLMs) with prompt engineering, fine-tuning of LLMs with domain-specific information, and development of an ad hoc rule-based expert system. A comparative analysis will assess their feasibility, limitations, and practical value for dam safety applications.
“The ability to improve predictive maintenance and anomaly detection in dams is essential, particularly as infrastructure ages,” said Professor Salazar. “DAMSHAI will help determine whether AI agents can complement traditional methods and reduce the workload on engineers while maintaining the reliability required for critical infrastructure safety” stated the expert.
Building on Established Expertise
The project builds upon decades of research by Professor Salazar’s team in dam safety and machine learning applications in civil engineering. It follows directly from DOLMEN, a recently completed CIMNE project that developed hybrid predictive models combining physics-based finite element methods (FEM) with data-driven machine learning approaches. DOLMEN successfully demonstrated how these combined methodologies could generate dynamic warning thresholds for dam safety monitoring.
The research team brings together expertise from across CIMNE, including contributions from renowned concrete durability expert and Senior Distinguished Researcher Prof. Carmen Andrade and Dr. Fernando Rastellini, among others.
Structured Approach to AI Integration
DAMSHAI is organized around four main tasks: generation of a comprehensive knowledge base incorporating monitoring data, FEM simulations, expert knowledge and technical reports; comparative analysis of different AI approaches; prototype development; and application to a real-world case study.
The AI agent will be trained using structured information that combines real monitoring data with numerical models, expert input, and technical literature. The project will address relevant challenges such as model accuracy, interpretability, and practical integration into existing safety assessment procedures. Particular attention will be given to arch dams, analysing different geometries and loading conditions to generate insights into dam behaviour patterns.
A Pioneering Approach: Assessing Practical Impact and Limitations
While internationally similar AI initiatives exist in areas such as construction planning, transport safety, and hydroelectric plant operation, DAMSHAI takes a novel approach to evaluating AI’s role in dam safety. The project will explore potential limitations including data availability, AI model reliability, user acceptance, and regulatory barriers that may affect the integration of AI-based tools into established safety assessment processes.
According to Prof. Salazar, results from DAMSHAI should “contribute to the ongoing debate on AI applications in civil engineering”, as they will provide “evidence-based assessment of the advantages and limitations of AI-based decision support tools for critical infrastructure”.
DAMSHAI is funded through the Knowledge Generation Programme (Programa para la Investigación y el Desarrollo Experimental) with grant PID2024-157828OB-I00, supported by MCIU and European FEDER funds.
