Human-Enhanced Loop Modeling (HELM): Agent-Based Finite Element Modeling of Concrete Bridge Barriers

Researchers introduce HELM, a human-agent collaborative framework that automates finite element modeling of concrete bridge barriers by decomposing complex tasks into verifiable checkpoints. The system improves autonomous modeling success rates from 20% to 75% by integrating AI agents with commercial FE software, addressing a critical gap in automating safety-critical infrastructure analysis.

The HELM framework addresses a fundamental bottleneck in computational engineering: finite element modeling remains largely manual despite its critical importance for infrastructure safety. Bridge barriers must withstand extreme lateral forces under standardized testing protocols like MASH TL-4 and TL-5, requiring high-fidelity nonlinear dynamic simulations. Traditional FE modeling demands specialized expertise, consuming significant engineering resources across geometry generation, boundary condition definition, and material assignment—tasks prone to human error when performed at scale.

This research demonstrates how structured human-in-the-loop automation can bridge the gap between fully autonomous AI and human expertise. By decomposing modeling workflows into discrete, visually verifiable checkpoints, the framework leverages AI agents' pattern recognition while maintaining human oversight at critical decision points. The 75% success rate represents substantial progress, though the identified failure modes—spatial reasoning and algebraic logic limitations—reveal where current language models still struggle with technical domain requirements.

The engineering industry stands to benefit significantly from this approach. Infrastructure asset owners face immense backlogs of aging bridge inventory requiring analysis and retrofit decisions. Automating FE modeling could accelerate safety assessments while reducing consulting costs and human error rates. The open-source release of agent design code democratizes access to these tools, enabling broader adoption across firms lacking large automation teams.

The framework's applicability extends beyond bridges to other safety-critical infrastructure including dams, buildings, and transportation structures. Future development should focus on improving spatial reasoning capabilities and expanding material behavior models. Success in this domain could establish precedents for human-AI collaboration in other specialized technical fields requiring domain expertise and verification.

• →HELM increases autonomous FE modeling success rates to 75% through structured human-in-the-loop intervention and discrete verification checkpoints.
• →The framework integrates with commercial FE software (ANSYS and LS-PrePost), enabling adoption by existing engineering workflows without replacing current tools.
• →Spatial reasoning and algebraic logic remain primary AI limitations, indicating where human oversight provides the most critical value in technical automation.
• →Open-source release of the agent toolkit democratizes access to FE modeling automation, potentially accelerating infrastructure safety assessments industry-wide.
• →The approach establishes a replicable methodology for automating complex technical tasks requiring both AI capabilities and human expertise verification.