VBI-ChatGPT

A conversational front end for vehicle-bridge interaction analysis. Describe a bridge and a vehicle in plain English or Chinese; an agent builds the finite element model through a constrained set of tools, OpenSees solves it, and the results come back explained.

Checking whether the demo is online…

What it does

Builds a 2D beam bridge in OpenSeesPy with arbitrary supports and element-level damage.
Drives a vehicle across it — a quarter car, or a multi-axle truck with bounce and pitch — with full iterative coupling.
Recovers the bridge's natural frequencies from the vehicle's own acceleration, which is the premise of drive-by monitoring.

How it is built

The agent cannot touch the model directly. It reads state through inspection tools and changes it through editing tools, so an invented parameter becomes a validation error rather than a wrong answer presented confidently.

Nothing is solved until you have seen it: the schematic goes on screen for review, and approval is a hash the agent has no way to set itself.

Validation

The solver is checked against closed-form results: modal frequencies of a simply supported span, static midspan deflection, and the exact fourfold frequency shift when a midspan support is added.

Built with FastAPI, OpenSeesPy and NumPy. Interface in English and 简体中文.